This week, I’m doing something different with my blog and my podcast. I’ve compiled a list of the most common questions I get about AI when I’m teaching courses and leading professional development on the topic.

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 Spotify.

 

What would you recommend, in terms of school-wide or district-wide policies?

When ChatGPT first hit the scene, schools scrambled to develop a schoolwide AI usage policy with clear boundaries that students could follow. You can use generative AI to get feedback on your work but not to solve problems. You can use it to edit but not to write. You can use it as a study tool but not as a tool for summarizing information. You get the idea.

Here’s the problem. AI integration is deeply contextual. You might have a science course where an AI video editing tool saves time on “jump cuts” but a film class where proper editing is a key skill you want students to master. You might have a health course where students use AI to design an app but a computer science course where you need to learn to code by hand. In other words, the learning outcomes should drive the AI usage.

This means that we will need to move away from a one-size-fits-all model for AI integration and into something much more contextual, where we will move along a sliding scale. That’s why I created this AI integration continuum.

As educators, we can then be clear about when and how students use AI in our classes. Some teachers use a stop light approach (green means you can use AI, yellow is use it cautiously, red is you can’t use it at all) but even that doesn’t capture the nuanced reality that you will need to move back and forth on this continuum throughout a single project or assignment. The goal, then is to be as clear and explicit with students as possible.

Also, be sure to follow state and national policies. Here in the US, this means following COPPA, CIPA, FERPA, and other data privacy policies.

 

Should we use AI checkers?

AI checkers are a bad idea if you’re using them to catch cheating. Despite the claims otherwise, every time I have field tested one, the results are in the 80-90% range, which means 2-4 students in a class of 20 (my class sizes were never that small) will either be getting away with cheating or they’ll be falsely accused. I’ve used The Federalist papers as a way to test certain AI detectors only to find that 40% of Hamilton’s work was AI-generated. Now, unless he had a DeLoreon and a Flux Capacitor, I’m doubting that Hamilton used AI.

So, what’s going on? The AI detection software looks for patterns like black and white thinking, variance in transition words, consistent verb tenses, and clauses that break sentences into threes. Turns out this fits Hamilton really well. And at the same time, that would have been seen as quality writing. Indeed, it still should be considered a sign of complex thinking in writing.

But when use AI detection tools to spot AI, we essentially use AI to catch AI, which feels a bit like Blade Runner. But often, the students suffer from the results of these accusations. It creates mistrust between teachers and students and that’s something I never want to risk.

This isn’t entirely new. I had a high school teacher accuse me of plagiarism because she thought my writing sounded too advanced for a freshman. She apologized in a half-measure (not saying “I’m sorry” but saying “you have to understand that your writing was really advanced and I made a bad assumption”). I literally never spoke to her for the entire year. I was so hurt by the accusation. But that was one small accusation. For some students, this could lead to a loss of scholarships and for the school district, it could be a massive lawsuit.

Certain students are more likely than others to get falsely accused for using AI. High achieving students who write with precision and consistency while also employing a large vocabulary are more likely than others to get flagged for using AI. Autistic students, who tend to write a bit more formulaically and use consistent verb tenses, are also more likely to be falsely accused for plagiarism. Finally, ELL students who look at academic work as exemplars and mimic that style, are super vulnerable to accusations.

This doesn’t mean that we can’t use AI checkers at all. It’s just that we need to rethink their role in the writing process. AI checkers can be a great diagnostic tool to tell a student when to vary phrasing or to add more voice or style. In other words, when is the writing becoming derivative or formulaic? It becomes a learning opportunity for revision. This creates soft accountability while avoiding a rush to judgment or the potential for false accusations.

 

How do we keep kids from cheating?

This has been one of the most common questions I have gotten from the very get-go with generative AI and the fact that it hasn’t let up suggests that it’s still a very valid concern. I’ve seen so-called “experts” scold teachers for asking this question. Simply craft more engaging lessons and kids won’t want to cheat. Unfortunately, that’s not a realistic view of reality.

Students cheat for so many reasons: they’re busy in overscheduled lives and they’re looking for a shortcut, they’re worried about scoring well on tests (and getting into the right university) and they don’t trust their academic knowledge, they hate the content and feel stuck in a system that forces them to learn things they find irrelevant.

We can be proactive in addressing each of these issues. But we can also take AI-resistant approaches to our instruction. This might involve going in-person and lo-fi. There has rightfully been a major backlash against screentime in schools and going screen-less helps. However, we can also craft assignments that require the type of thinking that AI struggles with: real-time inputs, contextual thinking, and intrapersonal knowledge. I’ve written before about how we can rewrite our writing prompts to be AI-resistant. 

But we can also use AI ethically and wisely in a way that leads to deeper learning. The goal here is a co-creation model, where students don’t fall into the trap of cognitive atrophy (where AI does the thinking for them).

A co-creation model might mean that you start out by engaging in a question and answer with an AI chatbot. From there, you engage in your own research. You ask clarifying questions to AI. You might ask for a list of resources to check out and then you do your research on your own. Eventually, you create your own outline. You then ask the chatbot to create an outline and you compare the two. As you write, you mostly write from scratch but you might also ask the AI agent to copy your style and then you heavily edit what it writes. Afterward, you ask the chatbot to play the role of an avatar for multiple readers and provide feedback. Note how this approach actively avoids cognitive atrophy.

 

How do we avoid cognitive atrophy?

I’ve hinted at this already but one of the biggest drawbacks to technology is the tendency toward cognitive atrophy. Just as a muscle atrophies when it’s not used, our minds experience the same process (though it’s a bit more complicated from a neuroscience perspective) mentally when we outsource cognition to machines.

I’ve experienced this myself with maps. I memorized so many streets and landmarks in Phoenix, AZ because I had to. In college, I would print a map off from MapQuest and navigate the streets like a pirate in a Toyota Tercel. Fast forward to my life in Salem, Oregon. I depend on Apple Maps and Google Maps (I’m not adventurous enough for Waze). I struggle to remember street names and I have lost the ability to navigate a city from memory.

We all experience cognitive atrophy. As humans, we lost the ability to memorize massive amounts of text due to the printing press. Our attention spans took a huge collective hit when we embraced the telegraph. But we also gained new thinking skills that incorporated technology. So, on some level, it’s about the trade offs. What are we okay losing? What do we gain with new technology?

So, when it comes to learning, we need to ask, “What are the outcomes that students need to achieve and how do we use technology to help them reach their goals?” There are times when we will not use technology at all (AI-resistant) and other times when we use it to deepen the learning and still other ways in which the technology will push us to redefine the learning outcomes itself (in the same way that the graphing calculator has transformed Calculus as a subject area).

On a basic level, this often means slowing down the prompt engineering process and avoiding the dangers of automaticity. This is one of the reasons I developed the FACTS Cycle for prompt engineering. My goal was to help students avoid the dangers of cognitive atrophy. Here’s how it works:

 

I want to give a small example of a way students might use AI as a study tool that avoids cognitive atrophy. In the first step, students submit their work for the previous unit of study. During this phase, they also make a t-chart of what they believe they know and what they don’t know. Then the AI chatbot does the same thing. It provides a proficiency level for what the students know and don’t know.

From there, they compare their chart to the AI analysis. They can then focus on an area where they need help. They ask for an explanation, written in simple language, from the chatbot. They then ask follow-up questions. Next, they can engage in skill practice and then the AI tests them on what they know, getting progressively harder as they go.

 

At what age should kids start using AI?

Neil Postman was a technological historian and media critic who had a theory that our concept of adolescence is directly tied to literacy. The notion of teenage years and adulthood extending into later teen years had to do with the complexity of literacy and the fact that it took years to master. He warned that when kids use technologies designed for adults, we see a “disappearance of childhood,” where children act adult too early but also adults act childish too late in life.

I don’t entirely agree with Postman’s theory but . . . there is something to be said of the danger in having younger children using a tool designed for adults (not just in terms of safety or sexually inappropriate content). I’m very concerned with children forming bonds with fictional algorithms. I’m worried about them learning to communicate with chabots that are too positive. And I just think young kids need to be limiting screentime in general. So, I would recommend using AI in limited ways and often on the teacher side rather than the student side when it comes to elementary school.

 

What unintended consequences have you seen, negative or positive with AI?

One of the core ideas of The AI Roadmap is that we cannot predict how technology will change our world. We have to fight against the darker elements, embrace the benefits, and remain curious in the moment. I think that curiosity piece is huge. We need to be asking questions first and then getting more analytical and judgemental afterward. This helps us remain intellectually humble and avoid some of the extreme black and white thinking around AI (both the technophilia on one hand and the moral panic on the other hand).

However, a few years into it, we can already admit that there are some unintended consequences. First, it has made our society much more skeptical. But this has meant that I am constantly having to be more casual and add more personal details so that my voice sounds more distinct and folks don’t accuse me of using AI in writing. I hate, for example, that I’m not allowed to use the em dash anymore. That was my go to!

Second, I think there’s already a downside that we are seeing of AI in focusing on rote repetition while humans do the relational, contextual, problem-solving elements. Let’s let the machines do the repetitive work and we can handle the novel work. That sounds great in theory but if you look at call centers, you’ll see the problem. AI answers the simple questions but then humans are stuck with cognitively demanding, emotionally loaded problems that they have to solve constantly. They don’t get a break by answering “simple questions” and then tackling hard ones. It’s like being forced to solve the hardest math problems that feel impossible while never getting a simple problem – and then being yelled at the whole time from an actual human.

The third unintended consequence is almost the opposite of this. I worry about how often we use AI to write unpleasant emails and essentially do the unpleasant emotional labor we don’t feel like doing. That’s a huge issue in terms of our character and our authenticity and our empathy.

I also worry that we engineered AI to be so pro-social that it’s being “too nice” and failing to correct bad ideas. I worry about a generation raised on chatbots who tell them that everything is a great idea. It’s a bit like living in the world of the Lego movie. Everything is awesome. I have known people who became convinced to pursue really bad ideas because the AI chatbots were far too quick to agree with them. Which leads to my next concern . . .

I worry about the number of children who are developing relationships (including romantic relationships) with chatbots. I brought this concern up in my book and I definitely feel that people warned us about it. But it’s happening faster than most folks had anticipated.

 

What do you wish you had included in The AI Roadmap?

There are certain tools I developed for AI that I introduced in my blog (like the FACTS Cycle for prompt engineering) that would have gone well in the book. I think the AI continuum is a similar type of tool that would have made a lot of sense in the book. When I consider the last question I answered, about the unintended consequences, I wish I had addressed the downside of AI being so pro-social.

However, I am happy with the work I produced. I released that book at a time when most AI books focused on “How to” elements. I wanted to create something a bit more evergreen and focused on the “big ideas” relating to AI instead of the granular elements of “here’s how you do ______.” Also, many of the books  were written by AI evangelists and I’m proud of the fact that I was deliberately more skeptical and nuanced.

 

What about the environmental impact of AI? 

This is actually very mixed. In terms of data centers, the type that use the most energy are actually not generative AI but media streaming (though AI video creation comes close). I’m not going to stop watching Bake Off or rewatching The Good Place or Somebody Somewhere or Schitt’s Creek.

I actually had a long talk recently with a group of environmental scientists and they said that the environmental impact is more nuanced than people realize. There are huge efficiencies gained in areas like recycling (where sorting is now realistic), supply chains, crop yields (and reduced pesticide and water usage) but there is also a big upfront energy and water cost to training data. My friend Angela Watson actually has a great episode on the myths and realities around AI and the environment on her podcast.