In short:
What are you assessing students on(standards/rubrics)? - AI should not be used for this.
What aren't you assessing students on? - AI can be used for this.
If the focus is to get students to demonstrate skills/learning/growth in something, then the demonstration should be what AI can't help with.
Does that mean that all graded assignments should be done without AI assistance? Not necessarily.
Take the following example:
I teach a Web Development course. The course focuses on web coding skills in HTML and CSS, not on content creation. But, in order to write a webpage students need content to add to it. I don't care where that content comes from.
They can use Lorem Ipsum text if they want (a common practice in web mockups).
They can copy content from a wikipedia page.
They can borrow images from online.
In both of these examples, the borrowed content should always be sited (good educational practice) and the webpage should not be published because students have not been given permission to use this content. The content is purely to help in the demonstration of the skills they've learned.
Can AI be used instead of using Lorem Ipsum text or borrowing images? Absolutely! I'm not grading students on the content, so I don't care where the content comes from.
Can students use AI to write their code for them? Absolutely NOT! That would be cheating as I'm needing to see that students can do the coding.
Can AI be used to coach students through the code? That depends on the nature of the assignment. If this is an end-of-unit assessment when I'm expecting students to know how to do this, then no. But, if this is an assignment that students are using to still learn the process, I don't mind as I'd be doing the same thing if I was standing behind the student walking them through the process. But the emphasis here is in coaching, not doing the work. The AI would need to ask leading questions or demonstrate general chunks of code as examples without straight up giving them the full answer the student could copy and paste in.
If you think about your content area, you can probably think of ways AI could be used without breaching this line.
Conversations with AI in a Spanish class.
Website creation for a business class (where writing the website is not part of the graded criteria).
Topic ideas for a creative writing class.
Conversations with a historical figure in a history class, then writing a reflection of what you learned (student writes the reflection, not the AI)
Recipe Generator for a culinary class (where the student focus is on demonstrating cooking skills, not creating recipes)
Cheating with AI is too easy in The Age of AI. As an educator with 150-200 students per semester, how do you assess every assignment and ensure that students didn't cheat? It's a nearly impossible task. If AI can do most of our assignments, maybe we need to rethink what our assignments look like.
Skill21.org's Durable page helps educator's come up with assessments that are hard for AI to solve. They also help to rate the durability for existing assignments. You share an assignment with AI and use one of the Skill21 Durable prompts to ask AI to rate it for durability. It provides the rating along with feedback explaining what aspects of the assignment are easy for AI to complete and which ones are more "durable". There are multiple prompts provided to help come up with more AI-durable assessments.
Ditch That Textbook - AI in the Classroom: What's cheating? What's ok? (2023) - This article addresses different levels of AI use and even acknowledges the big gray-area where the same use of AI could be appropriate or inappropriate in different courses. Again, this all depends on the standards being addressed.