Talk on Generative AI in the Computing Classroom: Symptoms and (Possible) Antidotes

Talk

Generative AI in the Computing Classroom: Symptoms and (Possible) Antidotes

 

Abstract

Generative AI has begun to have profound effects on the delivery of many if not all computing courses. It is no longer only about what an educator does or allows, but what an educator doesn’t do or allow – and often why, or why not. Programming courses in particular are under pressure due to the proficiency of many GenAI models in working with natural language and code coupled with the myriad known issues that arise when teaching programming, particularly to large and diverse groups. There are going to be challenges for sure, but evidence is starting to emerge that there are many benefits to working with GenAI in programming classrooms. How to most effectively leverage GenAI in the programming classroom is still an open question. However, some valiant educators have embraced the idea and are actively trying various approaches and sharing their experiences. Many of these changes however, are just scratching the surface of the changes to come.
It is possible that GenAI will affect classroom dynamics and the very nature of our currently common pedagogical approaches to teaching computing. We can almost assume that if students use GenAI while learning programming with no guardrails in place, it will not result in an improvement in our current state. Copying a problem into a GenAI and having the AI produce source code is not going to be most effective approach. Somehow GenAI needs to be leveraged.
We have developed a new type of programming problem called Prompt Problems. Prompt Problems present the student with a visual representation of a problem. The student’s task is to prompt a GenAI tool which creates code that should solve the problem. If it doesn’t, the student can edit the prompt. We have also developed a tool called Promptly. Promptly presents prompt problems, and also runs the AI-generated code through several test cases. If the code does not pass all test cases, the student can edit their prompt and try again. The design (ideally) encourages students to specify and decompose the problem, read the code generated, compare it with the test cases to discern why it is failing (if it is), and then update their prompt accordingly. As we will discuss, there are two broad types of impact possible when using tools such as Promptly. The first is effects on classroom dynamics, and the second is better? worse? learning outcomes for students. Promptly is currently in trial on several continents in English-, Portuguese-, and Arabic-language classrooms. The talk will include a live demonstration and invitation to use Promptly.

 

Bio

Brett is interested in how humans learn to program and how they perceive this process. He is fascinated by the interaction between humans and computers, exemplified by his obsession with programming error messages and what AI has to do with them. He is far from alone in his belief that that generative AI will dramatically change the way programming is taught and learned and is keen to try to keep up with the seemingly non-stop acceleration of the capabilities of AI. He is not sure if he is surprised or not that LLMs have offered yet another parallel between programming and natural languages, in that LLMs have demonstrated similar capabilities in both domains through very similar mechanisms.

 

DBLP: dblp.org

Google Scholar: scholar.google.com