Back to History

Fields Changed

Registration

Field Before After
Trial Title Barriers to Retraining Barriers to retraining: Identifying motivation as a factor in the rural-urban AI-divide
Abstract We implement an online economic experiment to examine workers’ willingness to learn new skills and to identify potential barriers to retraining. The study is conducted via CloudResearch Connect with 500 participants targeted from states which lead in both forest cover and rurality - Maine, Vermont, West Virginia, New Hampshire and Alabama - to allow us to compare the behaviors of rural, suburban, and urban participants. Participants first engage in a set of tasks in which they encode words using a Caesar cipher. They are then offered the opportunity to engage in a new, more difficult task that involves the acquisition of a new skill, coding in Python. Finally, participants answer a survey about AI attitudes and perceptions as well as answer sociodemographic questions. This work attempts to better understand how new technology (AI) impacts the existing workforce and whether rurality impacts technology adoption. We also investigate job loss expectations to AI, main concerns of AI (employment impacts, forced retraining, environmental), and perceived skills mismatch. A rural-urban digital divide persists in the United States, despite significant efforts and successes in closing the gap, exacerbating uneven access to education, employment, healthcare, and innovation. As artificial intelligence (AI) has become a larger factor in our daily lives, this digital divide has spilled over into a rural-urban AI-divide, with urban adults of working age using AI at twice the rate of their rural counterparts. But the AI-divide may not be fully explained by differential access. This work uses an experimental design with random assignment to identify motivational aspects of the rural-urban AI-divide. In the study, participants engage in two rounds of an online task. In the first round, participants learn how to use a Caesar cipher to encode a word. In the second round, participants are asked if they would like to continue to perform the Caesar cipher task or learn a new encoding skill and use this skill to complete a task that will earn them a higher piece rate per correctly answered question. The information provided about the new skill differs by randomly assigned experimental condition. In the no information condition, participants are only told they will learn a new encoding skill. In the coding skill condition, participants are told they will learn how to use computer programming for encoding. In the AI skill condition, participants are told they will learn an AI-related skill for encoding. In all three conditions, the skill and tasks are identical – participants learn how to use a simple Python script to encode words with a Caesar cipher. We also implement a version of this experiment that uses the same three conditions but does not allow participants to choose the task they complete in the second round. After completing the second round, participants are asked to complete a survey about their AI experiences, attitudes, and beliefs. Finally, they answer a few sociodemographic questions. This work explores whether take-up of the new skill in Round 2 differs by experimental condition and by rural residence. Our design also allows us to determine whether engaging with the AI-related skill alters perceptions of the benefits of AI and confidence in AI-skill ability.
Trial Start Date May 26, 2026 June 14, 2026
Trial End Date June 09, 2026 June 26, 2026
Last Published May 11, 2026 09:06 AM June 13, 2026 02:36 AM
Intervention (Public) The primary intervention in this study is the introduction of a learning opportunity, with or without a piece rate bonus. The primary intervention in this study is the introduction of an opportunity to learn a new skill with differing amounts of information about the skill provided to the participant. Additionally, in one version of the study participants get to choose whether to engage in the new task, while in the other version participants do not choose.
Intervention Start Date May 26, 2026 June 14, 2026
Intervention End Date June 09, 2026 June 26, 2026
Primary Outcomes (End Points) Outcome 1: Acceptance of new learning opportunity Outcome 2: Completion of new learning opportunity Outcome 3: Number of tasks completed correctly Outcome 1: Acceptance of new learning opportunity Outcome 2: AI attitudes/confidence Outcome 3: Interest in learning new skill
Experimental Design (Public) Participant Recruitment: The online survey experiment will be conducted via CloudResearch Connect. We intend to include 500 participants targeted from states which lead in both forest cover and rurality - Maine, Vermont, West Virginia, New Hampshire and Alabama - to allow for comparisons across rural, suburban, and urban populations. Experimental Design Overview: STEP 1: Consent STEP 2: Participants complete a set of encoding tasks using a Caesar cipher. STEP 3: Participants are then asked if they would like to learn a new task - a brief intro to coding that will allow them to encode the words more quickly - or continue to encode manually using a new cipher. Half of the participants receive this learning opportunity with a higher piece rate and half are not offered higher compensation. STEP 4: Participants either continue to complete the initial task or the new task, depending on their choice STEP 5: Participants complete a survey re: AI perceptions, confidence, etc. Participant Recruitment: The online survey experiment will be conducted via CloudResearch Connect. We intend to include 500 participants targeted from states which lead in both forest cover and rurality - Maine, Vermont, West Virginia, New Hampshire and Alabama - to allow for comparisons across rural, suburban, and urban populations. Experimental Design Overview: STEP 1: Consent STEP 2: Participants complete a set of encoding tasks using a Caesar cipher. STEP 3: Participants are then asked if they would like to learn a new task. The information provided about the new skill differs by randomly assigned experimental condition. In the no information condition, participants are only told they will learn a new encoding skill. In the coding skill condition, participants are told they will learn how to use Python for encoding. In the AI skill condition, participants are told they will learn an AI-related skill for encoding. In all three conditions, the actual task is the same – participants learn how to use a simple Python script to encode words with a Caesar cipher. We also implement a version that uses the same conditions but does not allow participants to choose the task they complete in the second round. This allows us to understand how engaging in the new task affects attitudes toward AI regardless of willingness to engage in the new skill. STEP 4: Participants either continue to complete the initial task or the new task, depending on their choice STEP 5: Participants complete a survey re: AI perceptions, confidence, etc.
Randomization Method Randomization to the increased piece rate condition is performed by the Qualtrics software Randomization to experimental condition is performed by the Qualtrics software
Randomization Unit Individual, however, participants are clustered by state and rurality (rural or suburban or urban) before random assignment to the treatment within these clusters to ensure balanced numbers of treated participants in each subgroup Individual
Sample size (or number of clusters) by treatment arms 250 per treatment 100 per experimental condition in the choice version of the study 60 per experimental condition in the no choice version of the study
Power calculation: Minimum Detectable Effect Size for Main Outcomes See pre-analysis plan
Additional Keyword(s) AI, job retraining AI, job retraining, digital divide, AI-divide, rural-urban divide
Keyword(s) Behavior, Other, Post Conflict Behavior, Labor, Other
Back to top