Automation Vs. Augmentation: How Innovation Affects JobsIGIER VISITING STUDENT FILIPPO RASCHIA REPORTS ON THE SEMINAR BY DAVID AUTOR (MIT) ON THE EMERGENCE OF NEW WORK
Automation is a major new force in our society, and it exerts conflicting effects. On the one hand, automation erodes the importance of certain human skills. On the other hand, it complements human labor by letting new work tasks emerge. What is the overall effect on the types of job created and the overall structure of employment? This is the main question that professors David Autor, Anna Salomons and Bryan Seegmiller tried to answer in their recent paper “New Frontiers: The Origins and Content of New Work, 1940–2018”, which Autor (MIT) presented at a recent IGIER Policy Seminar.
The authors look at automation using two unique datasets. They first construct a database of new job tasks introduced between 1940 and 2018 by sourcing from volumes developed and used by U.S. Census Bureau employees to classify the free-text job descriptions supplied by Census respondents into occupation and industry categories in each decade. The second novel data source is a comprehensive, quantitative classification of the flow of patents over nine decades that identifies innovations which, on the one hand, complement occupational outputs (“augmentation”) and those that, on the other hand, substitute for labor-using occupational inputs (“automation”). These two conceptually distinct measures of innovation flows are developed using Natural Language Processing (NLP) tools: the first is constructed by calculating the overlap between patent texts and the micro-titles from the Census Alphabetical Index associated with each industry and occupation; the second captures overlap between the content of patents and the tasks that workers perform, as reported by the Dictionary of Occupational Titles.
Using these data, the authors show that, between 1940 and 1980, new worked emerged most strongly in middle-paid production and clerical occupations, while after 1980 new work arose disproportionately in high-paid technical, professional, and managerial jobs and, to a lesser extent, in low-paid personal and health services occupations. This divergence is reflected in the educational skew of new work, especially among jobs held by non-college workers: before 1980, new non-college work emerged where non-college workers were already concentrated, that is, in the middle of the occupational pay distribution; after 1980, new non-college work emerged in lower-paid health and personal services, and to a lesser degree in higher-paid sales and professional occupations, even as the stock of non-college work remained relatively concentrated in middle-skill occupations.
From 1980 forward, instead, the flow of new work employing college-educated workers skewed even further into the high-skill domain: rising sharply in professional and technical occupations and declining in clerical and administrative work. Hence, the emergence of new work and its locus have been the true driver of the current employment polarization.
The authors proceed by presenting a theoretical model of these trends. They list three forces which shape the endogenous creation of new job tasks and the elimination of old tasks: augmentation, which generates new labor-using job tasks; automation, which eliminates labor-using tasks; and shifts in consumer demand which affect task automation and new task creation by changing innovation incentives.
Next, they test new implications of their model using their data. Consistent with their approach, most of the contemporary employment is found in new job tasks added since 1940, and the changing locus of innovative activity predicts where new tasks emerge. Augmentation innovations strongly predict the locus of new task emergence, as measured by job titles, across occupations and over time. Automation innovations do not predict new title emergence, despite their positive correlation with augmentation innovations at the level of occupations.
The authors also find that exogenous demand shifts for occupational outputs matter too: they predict when and where new work emerges. To isolate negative demand shifts, the authors exploit an exogenous decline in industry-level manufacturing demand resulting from rising Chinese import competition over 1990–2018; to isolate positive demand shifts, they leverage changes in population age structure that indirectly affect employment through patterns of consumption. They find that increases in demand favoring a sector lead to the emergence of new work tasks, while conversely, adverse demand shocks retard the emergence of new work tasks in exposed sectors.
Lastly, they find robust evidence that labor demand shifts outward with new task creation and inward with task automation: employment and wage bills expand in occupations exposed to augmentation innovations and contract in occupations exposed to automation innovations. Bearing in mind that augmentation and automation innovations are positively correlated, these countervailing relationships with occupational labor demand are all the more striking.
by Filippo Raschia