I prefer CSV files because they're universally compatible with nearly all AI tools, including free-tier platforms like ChatGPT-4o, Claude, and open-source models. While tools like Microsoft Copilot can handle XLS/XLSX formats, some AI models don't. CSV simplifies data parsing and works well with Excel.
2. Randomizing Student IDs
Creating arbitrary student IDs protects student personally identifiable information. In Excel, you can use the formula =INT(ROUND(RAND(),4)*10000) function to generate unique 4-digit identifiers or create a three-letter code (e.g., AAA, AAB) and drag-fill the series. For example, I assigned codes like “AAD” to the same student in my source data, which has student information, and in my CSV, which doesn't. I use the random code to re-match IDs later. This approach balances privacy with practicality, a principle that applies equally to business contexts like anonymizing customer feedback or employee evaluations.
3. Manual Grading
The assignment was graded manually by my teaching assistant using a rubric I designed, which emphasized critical thinking and prompt quality. I instruct my teaching assistants not to use AI for grading because it is only appropriate as a tool for information feedback that supports learning rather than being used for assessments that determine grades. While AI-generated feedback helped students refine their work, final grades reflected human judgment to ensure fairness and accountability. This distinction mirrors business practices where AI might analyze sales proposals for improvements, but final decisions rest with managers.
While this example focuses on student feedback, the same principles apply to any scenario involving qualitative analysis of spreadsheet data—employee performance reviews, customer surveys, or project evaluations. Feel free to reach out if you’d like to discuss adapting these strategies to your specific context!
Thanks for the insightful article. As someone who has no exposure to using AI (yet) and PTT (non research), I had the following qns:
1. Why use CSV instead of Excel?
2. How do you randomize ID for students and then re match them correctly?
3. If the AI generated feedback was not used for grading - did you/TA grade submissions manually or was the assignment not graded for points?
Thanks.
Thank you for your thoughtful questions.
1. CSV vs. Excel
I prefer CSV files because they're universally compatible with nearly all AI tools, including free-tier platforms like ChatGPT-4o, Claude, and open-source models. While tools like Microsoft Copilot can handle XLS/XLSX formats, some AI models don't. CSV simplifies data parsing and works well with Excel.
2. Randomizing Student IDs
Creating arbitrary student IDs protects student personally identifiable information. In Excel, you can use the formula =INT(ROUND(RAND(),4)*10000) function to generate unique 4-digit identifiers or create a three-letter code (e.g., AAA, AAB) and drag-fill the series. For example, I assigned codes like “AAD” to the same student in my source data, which has student information, and in my CSV, which doesn't. I use the random code to re-match IDs later. This approach balances privacy with practicality, a principle that applies equally to business contexts like anonymizing customer feedback or employee evaluations.
3. Manual Grading
The assignment was graded manually by my teaching assistant using a rubric I designed, which emphasized critical thinking and prompt quality. I instruct my teaching assistants not to use AI for grading because it is only appropriate as a tool for information feedback that supports learning rather than being used for assessments that determine grades. While AI-generated feedback helped students refine their work, final grades reflected human judgment to ensure fairness and accountability. This distinction mirrors business practices where AI might analyze sales proposals for improvements, but final decisions rest with managers.
While this example focuses on student feedback, the same principles apply to any scenario involving qualitative analysis of spreadsheet data—employee performance reviews, customer surveys, or project evaluations. Feel free to reach out if you’d like to discuss adapting these strategies to your specific context!