Fig. 1: Participant recruitment flow chart.

All participants were recruited from an online data collection platform called Prolific. We screened 1,000 users for the following characteristics: USA resident, fluent English speaker, high approval rating on Prolific, no previous completion of any of our group’s experiments, balanced on sex. Participants were screened with the World Health Organization’s ASSIST, a questionnaire that allows for screening of use of substances such as alcohol, cannabis, tobacco, stimulants, inhalants and others. We also collected a lab-standard demographics survey to assess basic demographic features such as age, sex, level of education, income and race. From the screening surveys, we recruited cohorts for moderate to high alcohol and cannabis risk scores as identified by ASSIST (>3 for alcohol, >4 for cannabis). From the eligible candidates, we collected data for the slot machine game in 2 separate batches (over 2 days), resulting in 100 alcohol users and 94 cannabis users. Participants with substance use were additionally asked to complete further group-specific questionnaires (ADS and AUDIT for alcohol users, SDS for cannabis users). Finally, data from each group underwent a series of quality control checks. We screened out participants that failed the attention checks during the experiment, had less than 50% optimality of machine selected and minimal to no variability in the rating scores. A total of 68 participants were included in the alcohol group and 65 in the cannabis group.