How to create assignments in the age of AI?

Undergraduate students use ChatGPT massively. In the article “I am a student. You have no idea how much we are using ChatGPT.”, Owen K. Terry describes how undergraduate students rely largely on ChatGPT (The Chronicles of Higher Education, 2023). Terry, an undergraduate student at Columbia University, suggests that if education systems are to continue teaching students how to think, they need to shift to AI-proof assignments like oral exams and in-class writing.

Yes, many educators who are tired of facing plagiarism may agree that oral exams, also known as “viva” exams, can be the ultimate tool to force students to study. In fact, most universities and colleges in Asia, including top science and engineering schools, have oral exams included in their final assessments; however, only educators who have tried this understand how exhausting it can be. Interviewing every single student in a class, even if it is once a semester, is no joke.

Despite these challenges, some educators have found a way to make oral exams more manageable. At MIT’s Generative AI and Education Symposium, Dr. Eric Klopfer shared that several courses at MIT were already using teaching assistants (TAs) for oral exams (MIT Symposium, 2023). So, it may be possible, at other universities too, to give oral exams with the help of TAs.
The concept of in-class writing is not easy to implement either. Many students need to be “in the mood” to write creatively and forcing them to write at a specific time, within a period of 30 or 45 minutes, may not support all types of learners. However, it may be more feasible to have short writing sessions followed by discussions or other engaging classroom activities.

Is it possible for educators to facilitate students to think, create, collaborate, and learn without using extreme measures such as oral exams or in-class writing? If AI is to remain highly accessible for everyone, the answer must be yes.

Educators are taking two main approaches to combating students’ over-reliance on AI. The first is to focus on the ‘process' a student follows and not just their outcomes. This is so students are required to ‘show’ their effort. In fact, formative assessments are quite common in education.

The second approach is to raise the level of expectations so students are assessed based on the quality of their work, regardless of whether or not they use AI. These two directions are thoroughly studied in education research and are known to work well. We never pivoted to embrace these approaches because our outcome-oriented pedagogy is deeply ingrained in the education systems.

For instance, Cynthia J. Brame’s book Science Teaching Essentials, published in 2019, has several useful resources on creating assignments and exams that promote engagement, learning, and reflection (Cynthia, 2019).

Cynthia suggests two principles when designing assignments. First, assignments should promote "task-value". Task-value theory suggests that the value individuals place on a task influences their motivation and engagement. If students perceive a task as more valuable, they are more likely to approach the task with enthusiasm and persistence, leading to better learning outcomes. Her second principle is that assignments should promote "self-efficacy". Promoting self-efficacy involves helping students develop a belief in their own abilities to successfully complete tasks and achieve goals. If students see a task as doable and do it, they start developing the confidence to do harder tasks.

Therefore, enhancing our assessments relies on increasing students' perception of an assignment’s value and increasing their belief that they can effectively complete the task. Following are some strategies to create more effective assignments:

1. View students as producers of knowledge, not just consumers
We often think of students as consumers of knowledge, learning the key concepts and skills they need to move on in their chosen fields. We should instead shift to seeing them as producers of knowledge, as we do in our professional lives. This means assignments should include an open-ended problem and an authentic audience. Students should also have a degree of autonomy, choice, and control in the shaping of their project.

2. Work with students to scaffold the process
The idea of process scaffolding is to break an assignment down into layers so students submit their in-progress work, as well as their final outcome. While educators can easily create steps to complete a large project in class, involving students in the planning process can be highly engaging. Students can plan when they will do each step and discuss what resources they need, how they will review their work, and how to acquire feedback.

 
3. Scaffold cognitive skills
Assignments should be scaffolded so that students perform low-level cognitive tasks followed by metacognitive tasks. For example, a series of assignments may start with summarizing and analyzing, followed by critiquing, synthesizing, and reflecting. Providing students with opportunities to reflect on their work can enhance higher-order thinking skills. We encourage educators and students to use our tool, Process Feedback, to reflect on both writing and coding processes.

4. Let students think about the quality of their work by scaffolding evaluation
Educators should provide an opportunity for students to think about the quality of their work. We can ask students to spend part or all of a class period designing a rubric. This ensures that students have thought about the level of work they should be producing and that they are invested in the grading criteria. Specifically, educators can have students work in small groups to identify characteristics that should be examined in the grading process and identify markers of excellent, good, and mediocre performance for those characteristics. It is up to the educator to discuss and decide upon the final rubric.

Overall, in creating assignments, educators can benefit from focusing on the process, raising expectations, and creating engaging assignments based on task-value and self-efficacy principles.


Image credits: Subodh Dahal (not AI)