Innovative approaches and artificial intelligence can streamline and enhance the creation of formative assessment questions. Formative assessments, or assessments for learning, are ongoing evaluations used by teachers to monitor student learning, provide feedback, and guide instructional decisions, enhancing the overall learning process.

This article aims to explore a novel workflow that leverages the power of structured data, specifically spreadsheets, and the capabilities of ChatGPT’s Code Interpreter to generate customised classroom assessments. Take it as a prototype or proof of concept. Utilising a detailed spreadsheet to outline requirements offers a sophisticated approach to crafting questions aligned with various curricula, grades, subjects, and educational standards.

The journey from spreadsheet to the classroom involves multiple stages, from understanding requirements to question generation, quality assurance, and final output. It consists of a combination of technology and pedagogy.

The Workflow Process

The workflow process is organised into four main stages: Understanding the Requirements, Question Generation, Output Formats, and Quality Assurance. They create a seamless transition from defining educational needs to delivering customised assessments.

Understanding the Requirements

A structured spreadsheet guides the entire process. The spreadsheet is divided into various tabs, from questions alignment with standards, guidelines for crafting questions, and quality assurance criteria. It serves as a roadmap and a reference, ensuring the generated assessments are tailored to different curricula, grades, subjects, and topics. All requirements are inputted into the spreadsheet. This is where creativity meets precision, as questions are designed to align with specific standards, difficulty levels, and Bloom’s taxonomy. Multiple-choice questions, true/false inquiries, short answer challenges — the diversity of question types adds richness to the assessment, promoting a comprehensive evaluation of student’s understanding and skills. These can be based on the teacher’s requirements, e.g. short class quizzes, homework assignments or end-of-term tests.

Here is a run-through of how the spreadsheet is structured.

Question Generation

The spreadsheet is then uploaded to ChatGPT’s Code Interpreter model with a detailed prompt. Here’s the prompt:

Create School-Based Assessment Questions Using the Provided Spreadsheet

As an expert in schools-based diagnostic assessment, you are tasked with creating low-stakes tests and diagnostics for various subjects and grades. Please refer to the attached spreadsheet and follow the guidelines below to generate questions that align with the specified curriculum, standards, and quality criteria.

Curriculum: Refer to the “Questions” tab for the selected curriculum (e.g., CAPS) and align the questions with appropriate standards found in the “Standards Alignment” tab.

Grade, Subject, and Topic: Tailor the questions to the grade, subject, and topic specified in the “Questions” tab.

Type of Questions: Utilize the “Question Templates” tab to generate questions according to the selected type (e.g., Multiple Choice, Short Answer).

Bloom’s Level: Align the questions with the specified level of Bloom’s taxonomy, as guided by the “Guidelines” tab.

Difficulty Level: Adjust the complexity using the criteria in the “Guidelines” tab.

Number of Questions: Refer to the “Questions” tab for the specified number.

Inclusion of Images: Include or exclude images based on the selection in the “Questions” tab.

Output Format: Format according to the selected question format in the “Questions” tab and follow the “Formatting Guidelines” in the “Guidelines” tab.

Rubric and Answer Sheet: Include if specified in the “Questions” tab and align with the “Standards Alignment” tab if applicable.

Quality Assurance: Review for accuracy, relevance, and alignment, and apply the checklist or criteria from the “Quality Assurance” tab.

Please refer to the attached spreadsheet, which contains the necessary information across the “Questions,” “Guidelines,” “Question Templates,” “Standards Alignment,” and “Quality Assurance” tabs.

The user then engages with spreadsheet requirements via the chat function, checking the process is working as it should.

Quality Assurance

Each question undergoes a rigorous review to ensure accuracy, relevance, and alignment with educational objectives, as per the criteria inputted in the spreadsheet. From factual correctness to alignment with specific curriculum standards (e.g., CAPS), the process ensures the questions are engaging and educationally sound. Quality assurance includes automated checks for grammar and content relevance and manual reviews by educational experts. For example, algorithms may screen for syntax errors, while human reviewers ensure alignment with specific educational goals like critical thinking or problem-solving skills.

Output Formats

Finally, the questions are formatted according to the selected output formats, such as Word or Plain Text. Including rubrics and answer sheets adds another layer of sophistication, providing teachers with ready-to-use materials that can be directly implemented in the classroom.

Example of outputs from this process (note, these are zero draft, i.e. un-edited):

Grade 4 Coding & Robotics Assessment

Grade 7 Coding & Robotics Assessment

Potential Issues and Limitations

Interpretation Challenges

While the spreadsheet provides a structured and detailed guide, there may be instances where guidelines or standards are misinterpreted. The nuances of educational objectives and the complexity of subjects can lead to ambiguities, requiring careful review and human expertise to ensure accurate alignment.

Technical Limitations

The process involves converting data from a spreadsheet into various output formats, such as Word or Plain Text. This translation can sometimes result in formatting constraints, especially when dealing with complex structures or intricate question designs. The flexibility and aesthetics of the final documents may be affected, demanding additional manual adjustments.

Quality Assurance

Automating quality assurance is a challenging feat. Ensuring consistent alignment with educational standards, maintaining the integrity of Bloom’s taxonomy levels, and verifying the accuracy of information requires a combination of automation and manual review. This hybrid approach ensures quality but may introduce a layer of complexity and time investment.

Scalability

The approach excels in crafting tailored assessments, but challenges may arise when scaling to generate many questions across diverse subjects and grades. Balancing customisation with efficiency while maintaining the highest quality standards becomes a delicate dance that may require further optimisation and technological advancements. To overcome scalability challenges, techniques like parallel processing and clustering could be employed, as well as access to more robust hardware and software solutions.

While these challenges present hurdles, ongoing research and collaboration between educators, technologists, and subject-matter experts could lead to improvements. Regular updates, technological advancements, and user feedback will likely help refine and optimise this innovative approach.

For teachers in the classroom, this approach offers a potentially valuable tool that saves time, promotes quality, and allows for tailored assessments that resonate with their students’ unique needs. It’s a glimpse into the future of education, where technology serves as a facilitator, empowering educators to focus on what they do best: teach, inspire, and nurture minds.

More on AI and Assessment

• Teaching Assessment Professionals to Work with AI – Training strategies for AI integration
• Adapting the AI Assessment Scale – Evaluation framework for your context
• Formative Assessment Fundamentals – Core principles for effective assessment