Chatbots vs. Cheatbots

If you devote any time to the question of AI in schools, read Carl Hendrick’s blog post “Children of the Magenta Line.”

For Hendrick, AI in schools is a “learning-architecture problem,” because the way that some people set up (“architect”) the learning experience with AI doesn’t actually result in learning. Specifically, the ease with which novice learners can use AI defeats the learning experience.

To whet your appetite, here are three big ideas from Hendrick.

1| You cannot always just “look it up.”

I’m old enough to remember the introduction of laptops and LANs in schools. When Google came along, a certain type of person argued that we wouldn’t need to teach kids facts anymore, because they could just “look it up.” (The classic example always seemed to be the War of 1812—“Why do we still ask kids to memorize the details of the War of 1812 when they can just look it up?”) 

Well, if you need to engage in analysis or critical thinking about new data or information, you can’t just look it up. (Or, in this case, trust AI to tell you whether an analysis is correct). As Hendrick shows, these kinds of skills are not general purpose. They’re domain-specific and knowledge-dependent: “You can't connect the dots if you don’t have any dots.” 

To say it another way, the knowledge that you previously struggled to consolidate enables you to make sense of new knowledge. 

Learners must do the hard work of refining data and information into knowledge, which typically begins by connecting them to prior knowledge (eg, “What have I learned about other conflicts that is similar to or different from the War of 1812?”). If a chatbot hands the student a complete picture, then no refinement takes place and the student connects no dots. Meanwhile, they’ve cheated themselves out of a learning opportunity.

2| Beware “surrogation”: performance is not the same as learning.

We know that a student’s grades are not the same thing as their learning. Grades are the teacher’s attempt to measure the learning.

But how often have you heard things like, “She’s an A student”? Or “He got all 5s on his AP exams?” Or “The average SAT score for the class was 1300”?

In each case, we are replacing the measurement of learning with the learning itself. That’s called surrogation. 

Surrogation also happens when we conflate the performance of learning—eg, the student’s production of an essay—with the learning itself (in this case, critical thinking, writing, and editing skills).

Chatbots make it easier than ever to optimize for performance at the expense of learning. Imagine that a student prompts a chatbot to write an essay on the most important causes and consequences of the War of 1812 (obviously). If we are assessing that student’s learning based on the product created, we’re dealing with surrogation on steroids.

Hendrick points out that performance (eg, solving a problem) and learning from that experience are distinct, and can sometimes even work at cross purposes. If that sounds counterintuitive, think back to middle school math, when you might have memorized that in a right triangle, A² x B² = C². You could perform the calculation, which might have been all you cared about to get a good grade. And that preoccupation with performance might have blinded you to learning something about why the hypotenuse matters. 

(To be clear, I’m not excusing anti-motivational instruction and putting the onus on students to find the meaning in the hypotenuse. I’m only saying that if a student believes that the goal is performance, we shouldn’t be surprised if their learning takes a backseat.)

3| Define powerful learning first, then harness AI to that vision. 

I hope you can see from the previous two points that students’ use of AI in unguided learning experiences harms learning.

But AI is an arrival technology, and we are only going to see more of it—everywhere—over the next decade. So what do we do?

Hendrick’s answer is to engineer “desirable difficulty” rather than removing it. As it turns out, this is exactly the premise of the Middle States Declaration of Powerful Learning, a centerpiece of our Next Generation Accreditation protocol and our AI endorsements.

The Declaration of Powerful Learning sets the context for effective AI use in the classroom by:

• Grounding learning experiences in a strong sense of purpose (or at least salience—the eternal “Why do I need to know this?” question);

• Further motivating students by empowering them with scaffolded, appropriate degrees of agency; 

• And—Hendrick’s preoccupation—focusing on mastery of knowledge and skills.

Powerful learning doesn’t happen in an hour—it develops over time. That’s why at Middle States we use the analogy of “games-scrimmages-practices.” 

At your school, is AI a chatbot that fosters powerful learning, or a cheatbot that robs students of the chance to develop mastery through purposeful, agentic experiences? Let me know what you are seeing: tag me on LinkedIn or email me.