Teaching programming with generative AI

The widespread availability of generative AI tools has changed the way developers and students approach programming tasks. Tools like Copilot can rapidly produce simple, boilerplate code and this should allow developers to move on to more interesting parts of the programming. However, these tools are not perfect, and the code produced by generative AI tools often contains non-obvious bugs. This shifts the programming task from writing correct code to understanding and debugging machine-generated code. This has implications both for practising developers and students learning to program.

For students, the use of generative AI shifts the balance of learning from learning how to write code, to learning how to read and debug code, while also encouraging students to consider higher-level design considerations. In this topic, we examine how our practise of teaching programming should change to account for this new environment. This is a new research topic, working in conjuction with the Technology and Education research group.