Generative A.I. has created some very real challenges with academic integrity. Schools have been scrambling to create systems and policies that address the potential for cheating. In the past, I’ve written about how we might redefine the essay in an age of A.I. or how we might incorporate it into our creative projects. But it’s easy to miss how A.I. tools can help us differentiate instruction in a more traditional lesson. Generative A.I. can create supports and scaffolds in a way that is faster and more feasible than ever. In this week’s article and podcast, I share five ways we can leverage generative A.I. for scaffolds and supports.

Listen to the Podcast

If you enjoy this blog but you’d like to listen to it on the go, just click on the audio below or subscribe via iTunes/Apple Podcasts (ideal for iOS users) or Google Play and Stitcher (ideal for Android users).

 

Making Differentiation a Reality

We all have things that we can independently but other things that are impossible to do. Not to brag but I’m pretty good at loading a dishwasher. On the other hand, I can’t slam dunk a basketball. However, there are also certain things in a middle zone that you can’t do on your own quite yet, but you can accomplish them with a little help. You might need help from a teacher, a peer group, or a resource. With proper training, hours of practice, and a trampoline, I could eventually slam dunk a basketball.

Vygotsky described this middle space as the Zone of Proximal Development (ZPD). Here’s how ZPD works. At the center, you have the things you can do on your own. On the outside, you have the things you cannot do. But in this middle zone you have the Zone of Proximal Development, which are the things you can do with guidance and support.

 

In 1976, Jerome Bruner applied Vygotsky’s theory to the K-12 educational setting with the concept of scaffolding. Here, educators provide supports, called scaffolds, to help students master the learning. Then, like the scaffolds in a building, teachers pull back the scaffolds as students master the knowledge. At this point, the ZPD grows outward as students master new knowledge with new scaffolds. might be tutorials, videos, and visual aids.

Here’s where A.I. becomes a game-changer.  It could be as simple as an algorithm with “recommended tutorials” for students who are struggling. You might use generative A.I. to design scaffolds from scratch or you might take existing scaffolds and use A.I. to modify them. For something more intensive, students might use generative A.I. as a type of tutor with a back-and-forth question-and-answer. With A.I., it is easier than ever to differentiate instruction in a way that helps students move through the ZPD.

 

5 Ways to Use A.I. for Student Supports and Scaffolds

The following are five ways you might use generative A.I. for scaffolding and supports.

1. Easily Create Leveled Readers

When I taught social studies, I used to do rotating reading centers. Students would go from group to group and fill out a graphic organizer in a carousel activity. This process broke up reading into smaller chunks and incorporated movement into the process. Students would read a 3-4 paragraph case study, discuss it briefly, and add details to a graphic organizer. When the timer went off, they would move to the next station. I used to create 5-6 stations with all of these examples and then modify them for text complexity. I spent hours working on these leveled readers.

With a tool like ChatGPT, I could easily take those case studies and have the tool create multiple versions at different reading levels. Students can then access the complex content in simpler language. You can use the same process for primary sources, where students might have access to a challenging reading that they close read but then a simple language version that makes the text more accessible. Here, students start by reading the text at their grade level to make sure they understand it. From there, they can use close reading strategies to annotate the original primary source written at a higher reading level.

You might use this tool with a complex entry from Wikipedia and create 3-5 versions of an explanation about covalent bonds or 19th century imperialism.

We can take it a step further by having the A.I. create the initial informational text from scratch. Here, you provide an outline of key information and use the generative A.I. to create a non-fiction text for you. You then read through the text and modify it. Chances are, it will be  a little boring, so you’ll edit the text and punch it up with some humor, some bizarre examples, and a few things you think the A.I. missed. You might need to fact-check it as well. From there, the A.I. chatbot can create a leveled reading ranging from 3^rd Grade to 7^th Grade. Now every student can access the text at a level that matches their fluency level.

Notice that you are still the expert. You’re not turning this process over to the A.I. entirely. Instead, you’re using it to create the initial version that you then modify based on your content expertise and your knowledge of your students.

 

2. Use A.I. to Build Conceptual Understanding

Generative A.I. tools work really well when asking questions and follow-up questions. While there can be challenges with bias and inaccuracies, it can be a great tool (with guidance) to generate multiple examples. So, if a student is struggling to understand linear equations, they can use a chatbot to ask multiple questions about how linear relationships work and what they look like in our world. They might start out by asking for a definition and examples but then move into a question about finding a linear equation when looking at a graph or a table.

If they’re struggling to understand how World War I started, they might ask questions like, “How was militarism related to imperialism?” or “Why would one guy getting shot lead to a whole world at war?” Again, they can use prompts like, “Write this in a way a 12-year old could understand.”

The following is an example of what this looks like in math. Note that I’m showing how a high school student might use it in a Statistics class:

You can think of the chatbot as a personalized tutor, where students are able to chase their curiosity and ask multiple questions. Students can ask for examples and non-examples. They can make distinctions between ideas. Along the way, they builder a deeper conceptual understanding of the subject.

It’s important, though, that we recognize the limitations of generative A.I. for conceptual understanding. Chatbots can sometimes fail to understand a question. If a student doesn’t know how to craft a good prompt, they can ask a question that leads to an inaccurate answer. On occasion, chatbots have been known to generate answers that are entirely made up. Because they pull from a massive data set and use natural language processing, they might answer a question with factual inaccuracies. This is especially true when a student asks chatbots for recommended sources or citations. Finally, generative A.I. can contain biases. They’re pulling from a massive data set and this can perpetuate stereotypes. Students often view chatbots as unbiased and objective but they should treat their interactions with the same skepticism they would use when doing a Google search.

 

3. Breaking Down Complex Tasks

Sometimes students struggle with task analysis. When directions are too complicated, they get overwhelmed and hit cognitive overload. This is especially true for students who struggle with executive function. Generative A.I. has been a game-changer for me in helping students take a seemingly impossible task and break it down into smaller sub-tasks.

A few months ago, I demonstrated how to use ChatGPT to design individualized assessments. During a break in class, I showed some students how I use ChatGPT to help me break down my home improvement projects and set more realistic time deadlines. It’s similar to this process of using A.I. within project management.

A week later, a student came to me and said, “I took the syllabi from my upcoming classes and and copied and pasted the assignments and due dates from Canvas. I asked for estimations on time given how fast it usually takes me to read and write. Then I used it to set up a course plan. I had it modify my course plan based on my work schedule and my preferences. I’m a night owl, so I’m not going to get my work done in the morning. From there, I broke different assignments into subtasks with due dates. I copied and pasted it into a spreadsheet.” This student then created checklists to stay on track.

This process was a bit complex. I wouldn’t expect an elementary or middle school student to know how to break down an entire unit plan or syllabus into tasks and subtasks. However, this might be a process that special education teachers can use when students need help in task analysis. They might even model the process for students so that as students with learning differences grow older, they can use this process on their own. It becomes another way they self-advocate and the A.I. functions as an assistive technology.

 

4. Quickly Create Scaffolds for Neurodiverse Students

The previous example is just one way you might use A.I. to generate structures and scaffolds for neurodiverse students. Here are a few more ideas:

• Providing additional handouts to facilitate task-analysis and executive function
• Using A.I. to help schedule small groups
• Using A.I. speech recognition software as an assistive technology to help students with writing
• Using A.I. image generators to help students who need a more concrete example of what they are learning in class
• Designing targeted skill practice. For example, you might use a chatbot to generate word problems for students who struggle with 2-step equations, or you might use it to create a high-interest non-fiction text at a student’s reading with sample questions
• Using A.I. to modify assignments to reduce cognitive load (fewer steps) while encouraging students to still access the grade level content.
• Using A.I. to reduce the amount of work while still maintaining a high challenge level. For example, a student with dyscalculia might need fewer problems but can still master the math content at the same grade level.

None of these supports should replace the goals within an Individualized Education Plan (IEP). We don’t want to replace educators with algorithms. We can, however, use the A.I. as a starting place for designing more personalized scaffolds and supports. I’ve seen Special Education teachers use goals from IEPs and current assignments to design individualized scaffolds and supports. From there, these teachers take the A.I.-generated scaffolds and modify them based on their own knowledge of the students.

This fits into this concept of A.I. as “vanilla” and then modifying the vanilla to create something better:

Again, the goal would be to take something general and make it more specific based on your relational knowledge of your students. A.I. makes the scaffolding process faster but then you can go more methodically and intentionally to design something individual students will use.

 

5. Creating Language Scaffolds

Previously, we examined how to help provide supports for neurodiverse students. But what about students who are learning English as a speaker of another language (ELL, ESL, ESOL stu- dents)? We can use A.I. as an initial starting place for creating language supports. These include:

• Front-loading vocabulary: you can use A.I. to identify some of the Tier 2 and Tier 3 vocabulary that students might need to master. While you’ll still need to create a list of vocabulary yourself (and rely on student feedback) the A.I. can be a great starting place. I’ve found that certain chatbots do a great job defining vocabulary in simple terms and even coming up with example sentences. If you couple this with an A.I. image generator, you can save time in generating front-loaded vocabulary.
• Providing translation help: While it still works best to part- ner students with someone who is multilingual, A.I. translators have come a long way. The dynamic aspects of an A.I. bot allow students to interact with the content in their native language while also being exposed to content in English.
• Providing leveled sentence stems: This remains a weaker area for A.I. but I am noticing significant improvements in A.I.-generated sentence stems, sentence frames, and clozes.
• Using visuals within the project to help facilitate language development: As A.I.-generated visual art continues to improve, we can potentially create additional visuals that can aid with accessing English.
• Assess language proficiency: A.I. can work as a formative assessment tool by analyzing a student’s speech or writing. This can be particularly useful in assessing language learners who may not have access to a native speaker or who are learning in a remote setting.
• Language practice: Students can provide the A.I. chatbot with the directions to engage in a language role-playing conversation. They can set the purpose, location, and fictional person they want to A.I. to pretend to be. Then, they can practice English with the chatbot.

Notice that a teacher can begin with these A.I.-generated supports but then modify them to suit their context. Teachers might even invite students to help with this modification process. This then frees teachers up to pay attention to a students’ affective filter and finding ways to reduce fear and anxiety.

I’ve noticed that ELL teachers tend to spend a significant amount of time designing supports and scaffolds. Meanwhile, many ELL students in a non-ELL classroom fail to receive certain supports they need. If we can leverage A.I. to save time in designing scaffolds, we can help students access the content while improving in their language development.

Empowering Students to Use A.I. for Supports

Last semester, I had a student ask for the transcript from our class Zoom session. He used the chatbot to delete the time stamps and translate it to Spanish. As a dual language student, he likes using both languages as he wrestles with ideas and compares the transcription to his notes in both languages.

I share this story because every time there’s a new technology and people are scared about cheating, I always ask, “How are people using this to scaffold their own learning?”

In other words, how might an exceptional learner use this? How might an English Language Learner use this? How might someone who hasn’t had the same advantages use this? Because what might seem to some as a “chance to cheat” might be a game-changer for someone else.

As educators, we can empower students to self-select the scaffolds they use. So, while you might make modifications for specific students (like the previously mentioned checklists or modified assignments) you might also have a bank of different tutorials and scaffolds that students can access if they need additional help. These supports should be available to all students. This approach embodies the idea of Universal Design for Learning.

Ultimately, we want to design learning experiences that empower all students to access the skills and concepts they’ll need for the future. But we also want them to use A.I. ethically and wisely. When students are empowered to use A.I. to design learning supports, they develop a mindset of using A.I. in a way that goes beyond cheating and instead treats the technology as a powerful learning tool.

What About Productive Struggle?

I also want to share some nuance here. I worry about the decline of productive struggle in a world of A.I. With productive struggle, students take risks, experience setbacks, engage in problem-solving, persevere. Here, we encourage students to grapple with complex problems and engage in deep thinking rather than providing them with immediate answers or solutions. By experiencing productive struggle, learners develop resilience, critical thinking skills, creativity, and a deeper understanding of the subject matter.

Moreover, we tend to value things more when we experience some struggle. In behavioral economics, they have the IKEA Effect. The phenomenon was first described in a 2011 study conducted by Michael Norton, Daniel Mochon, and Dan Ariely. In their research, participants were asked to assemble a simple piece of IKEA furniture. They found that participants who had assembled the furniture themselves perceived it as having greater value and were willing to pay more for it compared to those who received pre-assembled furniture.

When people invest their time, effort, and skills in building something, they develop a sense of personal ownership and attachment to the end product. The effort and accomplishment associated with completing the task contribute to a sense of pride and satisfaction, further enhancing the perceived value. In other words, the time spent doing the work on your own led to a higher value of both the end product but also the process. While this isn’t the same as productive struggle, it points to the value we place on things where we had to engage in more of the creative process. When answers are cheap and instant, I worry that students won’t value the learning as much.

Or consider the role of confusion. There’s some fascinating research regarding science videos and confusion. Researchers tested two types of videos. The first were clear and concise explanations of science concepts. The second set of videos were confusing and required participants to make a prediction, discover errors, and wade into the confusion. In the end, participants preferred the simple videos but they retained more information with the confusing videos. Derek Muller has an excellent video about this.

As we think about helping students use A.I. wisely, one of the critical elements will be the balance between getting immediate help and embracing productive struggle. There are no easy answers. As educators, we will need to design systems that embrace this balance as we move forward.

Get the FREE eBook!

Subscribe to my newsletter and get the  A Beginner’s Guide to Artificial Intelligence in the Education. You can also check out other articles, videos, and podcasts in my AI for Education Hub.