AI, Automation, and Continuous Learning in Career Choices

2025-02-17

2025-12-31

The main dependent variable captures which of two alternatives a participant selects, with each alternative varying across four attributes. This variable is assigned a value of 1 for the chosen alternative and 0 for the unchosen one. Our focus is on how the four attributes—salary, GAI use, automation risk, and continuing education requirements—influence the decision between the two alternatives.

We estimate heterogeneous preferences among respondent subgroups. In particular, we are interested in heterogeneity by respondent characteristics, such as gender and educational path (VET vs general education). Further, we will run two-way interactions of the following attributes: i) automation risk and continuous education ii) automation risk and GAI-usage iii) GAI-usage and continuous education.

To analyze students' preferences for occupational attributes, we implement a discrete choice experiment into a yearly representative survey of Swiss students in their final years of compulsory education and early years of upper secondary education. This timing captures students before, during or after this important transition between education levels, which typically occurs around age 15. The students are asked about their preferences for i) the occupation's risk of automation, ii) GAI use at work, iii) continuing education needs and, iv) salary.

To analyze students' preferences for occupational attributes affected by GAI, we implTment a discrete choice experiment into a yearly representative survey of Swiss students in their final years of compulsory education and early years of upper secondary education. In the first part of the survey exclusively students who are about to transition into upper secondary education in the current year are asked about their future educational plans (study/apprenticeship). In the second part of the survey, using a discrete choice experiment, we ask the full sample of about 4500 students to imagine that they need to choose between two occupations which vary in four attributes, defined as follows:

1) The occupation's risk of automation (continuous, 3 levels): The risk that your job will be completely replaced by new technologies such as robots and artificial intelligence in the next 5 years.

2) Genarative AI usage (continuous, 3 levels): Proportion of work tasks that you perform with the help of generative artificial intelligence (e.g. ChatGPT).

3) Continuing education needs (2 levels): So that you can always do your tasks well, regular further training is also common outside of working hours, e.g. in new technologies or social skills.

4) Gross montly salary (continuous, 3 levels)

Each respondent will be randomly allocated to three choice situations. To optimally select a design of 12 choice sets of 4 blocks with 3 choice sets each, we use Ngene.

In the third part of the survey, all students are asked about their socioeconomic background as well as their preferences for specific work characteristics (e.g. working with people, with hands).

The primary outcomes of interest are the coefficients corresponding to our four attributes. As secondary outcomes we will analyze the two-way interaction effects between automation risk, GAI-use and continuing education needs, as well as heterogeneity by respondent characteristics, in particular gender and students’ educational pathway. With the dependent variable being the likelihood of an alternative being chosen, we estimate the previously mentioned coefficients using mixed logit models (MXL).

We test whether students' (dis-)preference for working with GAI vary with their preferences for doing specific tasks, that is i) working with people, and ii) working with hands.

Randomization is done by computer, before sending out the invites to the survey.

Individual

No

4500 students

13500: 4500 individuals times 3 choice sets each

1125 observations for each block of three choice sets (there are 12 different choice sets in total, each participant sees 3)