The Robots Won't Cause Massive Unemployment This Time, Either

The article argues that translation is a useful case for thinking about automation and work. Even when machine translation gets very good, translators still do other parts of the job, and the remaining human tasks can become more valuable when the rest is automated.

It uses a task-based and O-Ring view of jobs to show why automating many tasks does not make the last task less important. The text says this can lower costs, let translators handle more projects, and potentially increase demand for translation services. It then connects this pattern to earlier cases such as cotton picking, bank tellers, and radiologists, and to long-run U.S. labor data that do not show sustained large-scale technological unemployment.

Reading notes

#

• Translation is presented as a good case for thinking about the relationship between automation and employment, because technology already produces better drafts and, even so, translators are still hired.
• The text says jobs are bundles of tasks, and that automation usually covers only part of them, leaving tasks such as client contact, clarifying ambiguous passages, and checking gaps outside the software’s reach.
• The task-based approach is linked to economists such as Erik Brynjolfsson, Daniel Rock and Michael Kremer.
• The O-Ring model is used to show that, in certain jobs, an error in one task can compromise the entire result.
• In an example with ten tasks, automating nine of them greatly improves total performance, but the last task remains decisive.
• In this situation, the rational employer tends to value more the person capable of doing well the task that remained outside automation.
• The text states that this applies to translators, as in the Duolingo case cited, which still uses humans to maintain consistency of style and tone.
• Partial automation also lowers the cost of the service and allows the translator to handle more projects, instead of producing all the drafts alone.
• With lower prices, more demand may arise for other types of translation, such as websites, email campaigns, warranties and registration cards.
• The author says it is easier to see the destruction caused by technology than the creation of new forms of work.
• The text mentions mistaken predictions about the mechanical cotton gin, bank tellers and radiologists.
• The conclusion is that the broader pattern of recent history is the constant increase in demand for human labor alongside technological progress.
• The cited charts of labor force, total number of jobs, unemployment rate and employment among prime-age adults do not show large and lasting technological unemployment.
• The text ends by saying that the amount of employment does not guarantee uniform job quality, but that topic is left for another time.