Beliefs and attitudes on automation and AI

Each individuals will see a political statement encompassing a specific vision on automation and AI, except the pure control group.

2024-10-03

2024-10-23

In part 1, several beliefs related to automation and AI:

a) the number of robots in their country, the country that has or has recently installed more robots, the industrial sector where these robots were mostly installed both worldwide and in their country, how AI works (incentivized);
b) number of jobs robots have taken out or created;
c) number of jobs that will be lost or created because of AI, both in the aggregate and by skill requirements of the jobs;
d) opinions on how robots and AI may influence the job market.

In part 2, several policy preferences and the willingness to politically mobilize on the topic:

a) if policy-makers should intervene on automation and AI, and why;
b) several policy preferences, like minimum wage, stronger regulations, universal income, breaking up big tech monopolies, education policies, lower taxes on labour, tax credits for innovation, taxing robots, unemployment benefits;
c) willingness to publicly support a statement on automation and AI, either balanced, optimistic or pessimistic

From Change.org, we will record the number of signatures per petition.

In part 2, we will also consider:

a) chances of losing one's job in the next 5 years because of automation or AI;
b) trust in several institutions, as the government, political parties, trade unions and tech tycoons.

From Change.org, we will try to link each signature to a questionnaire by looking at the timing of the signature and the completion time of the questionnaires.

Part 1: misperceptions on automation and AI

After having collected basic information on demographics and voting intentions, we will acquire two crucial pieces of information: the occupation of the respondent and the industry were he works in. In this way we will be able to assess the exposure of his current or past job to automation. Furthermore, in our empirical analysis, by combining this information with other demographic and social characteristics, we will be able to instrument the current exposure with the expected one by using the approach in Anelli, Colantone and Stanig (2021).

Another relevant part of the introductory questionnaire will collect information on cultural worldviews, so to be able to classify individuals in terms of cultural types between individualist/collectivist and authoritative/liberal using the cultural cognition questionnaire (Kahan et al., 2013).

After having acquired this background information, we will elicit several beliefs related to automation and AI (outcomes a)-c)). Questions a) will be incentivized, as participants will get a bonus fee of $0.50 if they are able to correctly guess the answer to a randomly picked question among those in a). The correct answers will be based on the Institute for Robots statistics for 2022. At the end of this part, we will have open question d) on the respondent’s opinion on the impact of automation and AI on the labour market.

Part 2:

In this part, we explore the influence that visions on technology have on policy preferences and the willingness to mobilize in support of these preferences.

We introduce at the beginning of this part the political statement encompassing a specific vision on automation and AI. The facts within the articles will not change, but the predicted outcomes of automation and AI and the suggested policies will differ between a “techno-optimist”, “techno-pessimist” and a “balanced” spin. Each participant will only see one statement, except for the baseline where no statement is shown. The statements will be followed by control questions to ensure the participants have read and understood the statement.

Later, we will elicit policy preferences. They will be elicited with a) a question on if policy-makers should intervene on automation and AI together with an open question asking for arguments to justify the previous choice, b) radio buttons on the preferred policies among a list of policies from Johnson and Acemoglu (2023). The subsequent part will be on political mobilization, as respondents will be asked to sign an online petition on Change.org. They can choose one out of three petitions based on the statements or not to publicly support any petition.

At the end, we will have a final questionnaire where we will elicit the chances of losing one's own job, additional demographic and socio-economic information, trust in several institutions, risk attitudes, feelings towards technology, information on the personal use of technology. After the survey is over, respondents will be shown a link to the petition on Change.org that they declared they were willing to sign.

Not available

Randomization is implemented with a double-blind procedure. In our oTree code, each row of the dataset corresponds to a different treatment in sequential and repeated order (baseline, balanced, optimistic, pessimistic) without the recruiting company Cint being allowed to see the dataset. Respondents are recruited and re-directed to the website by Cint to fill a row of the dataset without any control by us.

Respondent

No

11,600 individuals across the three countries (US, Germany and Italy) recruited through Cint, representing a representative sample of these countries in terms of age, education and gender

11,600 individuals

2,900 individuals by treatment arm: 2,900 baseline, 2,900 balanced, 2,900 pessimistic, 2,900 optimistic