How Educators and Students Should Use Language Models
J.T. Torres and Adam Nemeroff from the Chronicle of Higher Education recently had an insightful article about how to use language models in education. The gist of the piece was that we shouldn't be thinking about language from the perspective of product and instead focus on process.
I love this idea because it gets to go beyond standalone AI systems or even tools and think about partnership. And once we are thinking about partnership, the question switches from whether to make use of these systems to how.
So how can we best bring language models into the workflow of education?
The authors suggest two main entry points: articulating the problem and critical evaluation of the output.
For the first, the issue is how well a student can describe exactly what they are trying to accomplish with enough detail to control the generation. The authors even suggest learning how to prompt by building prompts for each other. The idea is to use other students as a sounding board for whether a prompt is robust enough and, if not, how to improve it.
For the second, have students focus on fact-checking, readability, and argument structure once a model has produced something. Have them test out where the models work well and where they work badly.
By thinking about process and interaction, we can get students to think deeply about what they want to say, test out ideas on each other, and then develop the critical skills they need to evaluate not only their own text but the text that surrounds us daily.
It is our responsibility to make sure that students know that there's a tool in the world that they can use and get them thinking about how to use it. Not to just get answers, but to develop the ability to generate answers on their own and understand where there might be problems.
The goal is to get them away from thinking about language models as production tools. Instead, get them to think about how they can get a language model to help them to produce better. Stop thinking that we are building standalone systems. We're building intelligent systems that are going to be part of our environment that will work as partners with us.
Kristian Hammond
Bill and Cathy Osborn Professor of Computer Science
Director of the Center for Advancing Safety of Machine Intelligence (CASMI)
Director of the Master of Science in Artificial Intelligence (MSAI) Program