great deal of journalistic and academic attention has been focused on the strong growth in productivity in the U.S. economy. Between 1996 and 2003, productivity rose at a 3 percent annual rate, double the pace of the first half of the 1990s. Automation, frequently driven by advances in information technology (IT), has been one of the sources of this productivity growth. To take but one recent example, Atmos Energy, a Dallas, TX-based gas company, is automating its gas meter reading capabilities by using wireless technologies, a move that will allow it to reduce its staff by 225 employees over the next five years and thus attain significant increases in productivity.

It is natural to wonder if such automation-driven productivity enhancements can be sustained. After all, it seems like so many tasks and components of jobs are now automated. And yet there’s an argument to be made that we are still in the early stages of a new wave of automation, which will profoundly affect the economy and significantly contribute to the productivity growth over the next 10 to 15 years.

“There’s an argument to be made that we are still in the early stages of a new wave of automation, which will profoundly affect the economy and significantly contribute to the productivity growth over the next 10 to 15 years.”

Industrial automation goes back to the Industrial Revolution of the 18th century, when machines replaced physical labor on a massive scale. From the advent of the steam engine to the assembly line, work previously done by human hands came to be done by machines that could harness the power of water, steam, and, eventually electricity. In the past 25 years, automation transformed manufacturing as industrial robots replaced manual jobs in industries such as automobiles, computers, and telecommunication equipment. More recently, automation has been primarily driven by IT. The toll booth collectors who have lost their jobs to EZ-Pass technologies may be a harbinger of future trends. It is possible, for example, that many cashiers in department stores and supermarkets will soon lose their jobs because of the advancements of the Radio Frequency Identification (RFID) tag technologies.

Most of the jobs lost to automation have been routine production jobs, according to the job classification proposed by former Labor Secretary Robert Reich in his 1991 book The Work of Nations. Examples of these jobs, which are characterized by repetitiveness and structuredness, include assembly line workers, foremen, data processors, and toll collectors.

The next wave of automation will affect not only routine production workers, but also the better-paid and heretofore more secure group that Reich called symbolic-analytic workers – engineers, office and knowledge workers, managers, educators, and other groups of mind workers. Although few of these jobs will be eliminated completely, many of the more routine tasks in these jobs will be delegated to smart machines within the next 10-15 years, leading to major restructuring and consolidation.

Symbolic Analysts

I recently taught a course on Advanced Technologies for Business Applications at NYU Stern. The students, who were predominately part-time MBA students, were asked to describe what parts (if any) of their jobs or the jobs of their closest colleagues, could be automated within the next 10 to 15 years. All the students were symbolic-analytic workers according to Reich’s classification, and some of them worked in managerial positions. Based on about 30 student reports, an interesting picture emerged about the types of jobs that can be automated and the extent and scope of this automation.

In general, jobs can be classified along three dimensions. First, repetitiveness – for example, a salesperson repeatedly meeting with clients. Second, stability – a job that does not change over time. For example, a salesperson meeting with the same client, as opposed to meeting different clients. Third, structuredness – a job that can be described with a clear procedure, perhaps even expressed as an algorithm. For example, a salesperson can have a structured interaction with the client asking several standard questions and making several standard offerings of products. Alternatively, the interaction can be unstructured and open-ended.

Many jobs consist of several tasks, with each task characterized by the three dimensions of repetitiveness, stability and structuredness. Graphically, a job can be represented with a set of points in the three-dimensional space shown in Figure 1, where each point constitutes a particular task of the job.

The tasks that are closer to the origin in Figure 1 – i.e., those that are high on repetitiveness, stability, and structuredness – constitute the primary candidates for automation. For example, the task of a salesperson meeting with the same client over and over again and interacting with the client in a structured manner, asking the same set of questions and offering a simple array of services based on the answers, is a good candidate for automation by an intelligent software agent. Moreover, most of the routine production jobs that have been lost to automation rate highly on all of the three dimensions. In contrast, the tasks that are away from the origins on all three dimensions are the hardest to automate. For example, the task of a salesperson meeting with a different and ever-changing clientele and having unstructured open-ended discussions with them is very hard to automate.

If all the tasks of a given job can be automated, then the entire job can be eliminated. However, this is unlikely to occur for most of the symbolic-analytic jobs since most of them have some tasks that are ranked high along at least one dimension in Figure 1. Therefore, most of the symbolic-analytic jobs can be automated only partially (if at all) within the next 10-15 years.

Extent and Scope

One of the surprising outcomes of the student projects was the extent and scope of possible automations they identified for different types of jobs in diverse industries, including accounting, finance, healthcare, human resources, IT, marketing and sales.

For example, one type of a job a student described as already automated is that of the Client Accountant. This job is responsible for ensuring that all the client’s transactions settle properly, all funds are transferred, and all the account balances are reconciled with various parties involved in a transaction. It is a very routine and paper intensive job that rates very high on all three dimensions in Figure 1 (the point is close to the origin). Over the past few years, this job has been automated in the financial services and other industries. A single client accountant can now monitor the transaction activities of 10 times more accounts than was feasible in the past.

Another example of a job currently being partially automated in some companies is that of a Marketing Associate, who helps create a company’s responses to various Requests for Proposals (RFPs) or Request for Information (RFIs). One of the tasks for which Marketing Associates are responsible for is the collection, reviewing, and compiling of the account-related information (such as performance figures, market values, etc.) into a presentable format. It is a laborious, manual process involving running various reports, cutting and pasting information from Excel and Word documents, and eventually building a PowerPoint presentation. In many applications this process is structured, straightforward, and does not require much creativity. It also rates high on all three dimensions in Figure 1, and is a good candidate for automation. Some companies are currently trying to automate this task. However, that does not mean that the job of a Marketing Associate will be eliminated, since it also involves other tasks that are less routine and structured. Instead, Marketing Associate jobs are more likely to be consolidated and restructured by automating the tasks of responding to RFPs and RFIs and letting Marketing Associates focus on the more human-oriented parts of their jobs.

These two examples represent the simplest types of symbolic-analytic service jobs that are currently the primary targets for automation. The students also provided numerous examples of more advanced automation tasks. Currently, many business processes have already been partially automated by delegating some parts to machines and other parts to humans. Examples of such human-centered tasks include moving information from one system to another or checking the results returned from one part of the business process before initiating another. These human activities are often required because various systems may not “talk” to each other or may return questionable results that need to be inspected before the business process can continue. These activities usually constitute the leftovers from previous automation projects and comprise the hardest parts of these projects that were left un-automated for the reasons mentioned above. Naturally, they are primary candidates for new automation attempts using more recently developed information technologies.

The students also explored various other jobs that are significantly harder to automate, such as new product development, sales support, systems analysis, and project management, which all require significant advances in technologies before smart machines can perform these jobs. Although they claimed that such unstructured, non-repetitive, and evolving jobs are impossible to eliminate, the students identified various tasks within those jobs that could be automated within the next 10 to 15 years.

More Deep Blues

Although many findings in the student reports were quite unusual, they should not be very surprising upon further reflection. Consider the chess program Deep Blue, developed by IBM, which defeated the world champion Gary Kasparov in 1997. Or the projects attempting to automate the art of painting, writing poetry, and composing music, such as robotic painter Aaron, music-generating software EMI, and Kurzweil’s Cybernetic Poet, that are described in Ray Kurzweil’s book The Age of Spiritual Machines. Although these efforts are still in their infancy, it is quite possible that significant progress can be achieved in the next 10 to 15 years. And if such highly creative, unstructured, non-repetitive, and evolving tasks as playing chess, painting, composing music, and writing poetry can be automated, then significant portions of the work currently performed by symbolic-analytic workers can also be.

“If such highly creative, unstructured, non-repetitive and evolving tasks as playing chess, painting, composing music and writing poetry, can be automated, then significant portions of the work currently performed by symbolic-analytic workers can also be.”

Moreover, the low-hanging fruits are being picked right now, as is evidenced from such activities as automation of the client accounting and marketing associate functions described earlier. As another example, Lehman Brothers Inc. is currently automating payroll and other administrative functions. The main question is: How far will the IT industry be able to advance along the three dimensions of Figure 1 within the next 10 to 15 years?

The scope and extent of possible automation of the symbolic-analytic jobs described is possible only because of the development of advanced technologies that can enable these automation processes. It is these technologies that will propel the continued productivity enhancements in the coming decade. Many smart devices and technologies have been developed over the past few years, including smart homes, refrigerators, laundry machines, even tires. These are enabled by so-called smart software that monitors their behavior and drives and guides these devices. Meanwhile, numerous information technologies help remove the user from the loop from various business processes and thus make these processes more automated. These include Web services that help distributed computer systems interact among themselves and understand one another without any human intervention, workflow automation, and document analysis and processing technologies. Much human effort in the knowledge economy pertains to the processing of various multimedia documents containing text, images, video, and audio information. This labor-intensive activity is very difficult to automate because it involves natural language understanding and/or computer vision, which constitute two very hard areas of computer science. However, significant progress has been made in both of these areas over the past several years, and certain types of specialized textual documents and images can be analyzed by machines now.

ecent advancements in networking and wireless technologies will enable the development of new automation methods and new ways to redesign business processes. For example, RFID tag technologies might allow for the elimination of the check-out lines in the department stores and elimination of many cashier jobs. The tags would also enable automation of the business processes in the supply chains resulting in numerous efficiency improvements. And EZ-Pass-like technologies are certainly not limited to toll collection applications. The EZ-Pass concept – a person walking through a monitoring device that recognizes the provided service and automatically bills this person – could find numerous applications in all spheres of business within the next several years.

The integration of wireless, location-based (e.g., Global Positioning Systems, [GPS]) and Web services technologies constitutes a powerful combination that would enable numerous automation applications within the next 10 to 15 years. As an example, there should be no need for parking meter inspectors in the future. When a parking meter expires, and the car is still located in the parking spot, computer vision technologies could read the license plate of the car and the pertinent information for issuing a parking violation ticket could be wirelessly sent to the central office using Web services technologies.

Machine-to-machine interaction technologies, which facilitate direct interactions between the machines, currently include distributed systems, networking, Web services, and workflow technologies. However, more complicated and smarter machine interactions will be possible in the future by integrating other types of technologies into the mix, including some of the artificial intelligence-based technologies.

Some automation applications require formidable computing power. So far, the IT industry has met this challenge and continues to follow Moore’s Law – the notion that computing power can be doubled essentially every 18 months. That makes computation-intensive automation solutions more feasible. It is important for some of the automation activities that Moore’s Law continues to be followed in the future.

It is easy to be sanguine about the promise of new technologies, and frequently IT advocates and industry representatives paint a picture of unblemished progress when discussing innovation. However, the next wave of automation will have both positive and negative outcomes. It will have significant effects on productivity in terms of improved efficiencies and increased production speeds which will reduce costs. But these productivity improvements will have profound effects on the labor market, with many jobs and job categories being restructured, significantly reduced, or eliminated. Of course, job restructuring and elimination in some parts of the economy will result in job creations in other parts of the economy. Companies have learned over the last few decades that information technology can be a powerful competitive weapon that can significantly affect the economy and the society at large. To be able to respond properly to this coming wave of automation that will change not only routine production but also symbolic-analytic jobs, it is crucial to study and discuss the effects of this wave of automation before it affects us in profound ways.

Alexander Tuzhilin is associate professor of information systems at NYU Stern.