Skip to main content
EDSL assignments let students make study design choices explicit in code, run AI simulations, and compare findings with human evidence. Students can also collect human responses as the primary source of data. Start with first steps and the Run your first survey tutorial. The tutorial covers a first survey and test responses; the object and task guides extend it to agents, experimental conditions, survey logic, and model comparisons. Colab setup provides a notebook environment without a local installation.

​
Set up your course

Local construction and test-model exercises require no account. For hosted model execution or human data collection, students need Expected Parrot access. Contact [email protected] about academic support and usage credits for your class. Define the learning objectives, required deliverables, and limits on model usage or study size. Have students estimate costs and run a small pilot before scaling up a simulation.

​
Use a playbook for an assignment

A playbook gives students reusable instructions for a project or method. Specify the inputs students should provide, the decisions they must justify, and the evidence they should submit. Students can follow the same process while choosing different topics or datasets. For example, provide the following instructions for a message-comparison assignment:
Ask me for three advertising messages, a product description, and two audience definitions. Help me formulate a research question and design a short survey to compare the messages. Explain what each question measures. Let me review the survey and proposed models, then estimate the cost of a small AI pilot. After the pilot, compare responses by message and audience, identify limitations, and propose a human study to test the hypotheses. Produce a report, the underlying data, and reusable EDSL code. Label simulated responses clearly.
Require students to review and revise the design, explain their choices, and evaluate the evidence behind the final report. For a human research assignment, have students design a survey or interview guide and collect human responses as the primary data source.

​
Use the Research Agent alongside EDSL

  1. Start a Research Agent session with the assignment or playbook and project-specific inputs.
  2. Review the research plan and draft instrument. Use the Builder to inspect wording, response options, and logic.
  3. Run the planned study and inspect the resulting data. Distinguish human responses from simulated responses in every comparison.
  4. Ask for the EDSL code and check that the analysis matches the research question.
  5. Share the session and required study materials with the instructor using the session’s Share control.

​
Suggested deliverables

  • A research question and explanation of the study design.
  • The survey or interview guide, scenarios, and any AI respondent definitions.
  • Model selections and execution settings for simulations.
  • Data and analysis code supporting the findings.
  • A report distinguishing observations, simulated results, and hypotheses, with limitations and proposed next steps.
  • A shared session or notebook that lets the instructor inspect the workflow.

​
Discuss what the evidence supports

AI simulations can help students explore hypotheses, test instruments, and examine model behavior. They do not automatically measure the attitudes or behavior of real populations. Have students compare results across conditions and models and explain what human evidence would be needed to evaluate their conclusions. Human surveys and interviews are also complete research projects in their own right. Choose the method based on the course’s learning objectives and research question. For examples, see Research with Expected Parrot, the Run your first survey tutorial, and Research Agent resources. See Citation for acknowledging EDSL in student work.