Uncertain Discourse Concerning the Future of Automation

• In Event: Studying the Impact of Information Technology on Citizens

Abstract

As artificial intelligence (AI) moves from research labs to real-world applications, AI systems are impacting more and more workers today. Economists have found that the proliferation of information and communications technology in recent decades has hollowed out many middle-skilled jobs and increased socioeconomic inequality (Acemoglu and Autor, 2011; Autor, Levy and Murnane, 2003). But technologists, social scientists, and journalists are not certain how AI advances will impact workers in the upcoming decades. Some forecast a dramatic transformation in which many – if not most – jobs will become automated (Brynjolfsson and McAfee, 2014; Frey, 2019). Others contest these claims by arguing that these momentous predictions are unsupported by current labor trends, such as low productivity growth and relatively low unemployment rates (Benanav, 2020; Surowiecki, 2017). Furthermore, forecasts of the number of jobs at high risk of automation are highly uncertain, ranging between 3.4 million and 80 million in the U.S. over the next two decades (Winick, 2018). How do workers respond to these conflicting messages about the impact of advanced automation on their jobs? This paper seeks to answer this policy-relevant question by understanding how expert disagreements regarding the future of work shape workers’ beliefs and attitudes.

First, we used natural language processing to perform a content analysis of articles about automation in mainstream U.S. newspapers from 2015 to 2021. Our research shows that while articles initially emphasized job loss due to automation more than job gain due to automation, content about job gain increased after 2018. As a result, readers are now consuming more mixed messages about how automation will impact workers. In addition, our analysis shows that national newspapers, compared with local newspapers, cover sufficiently more content about automation. Because the readers of national newspapers tend to be wealthier and more highly educated, those who follow the discourse around automation may be more elite than the average worker.

Second, we plan to expand upon a pilot study (N = 203) that used a survey experiment to test how different types of messaging impact the public’s beliefs about workplace automation and policy preferences. In this survey experiment, we randomly assigned respondents to read news articles (that we composed) where experts agreed or disagreed about the future impact of advanced automation on employment. Results from this pilot study show that expert disagreement (compared with expert agreement) appears to make respondents more uncertain about the percentage of jobs and tasks that will be automated by 2035. Furthermore, expert disagreement does not seem to change respondents’ (point estimate) forecast of how likely their own jobs will become automated by 2035. Finally, expert disagreement seems to reduce support for some policies to address the negative consequences of advanced automation, such as funding job retraining programs. While these initial findings are promising, we plan to confirm them using a survey with 2,000 respondents.

Authors

• Baobao Zhang, Syracuse University

• Chloe Ahn, University of Pennsylvania