An AI Homework Game That Makes Students Find The Mistakes

Parents worried about students using artificial intelligence to do their homework can try turning the technology on its head.

Instead of asking a chatbot for the right answer, ask it to be wrong.

The result can be an AI scavenger hunt in which a student uses what he or she learned in class to find deliberate factual errors.

The activity can work with almost any subject.

If a student has been studying the Missouri Compromise, for example, ask a chatbot to write a short, age-appropriate passage about the topic containing exactly three deliberate factual errors.

Or try photosynthesis, the causes of the Dust Bowl, the solar system, a book being studied in English class, or another topic covered that week.

Have the chatbot produce a separate answer key identifying and explaining the three errors.

The parent keeps the answer key.

The student gets only the flawed passage.

Then comes the important part.

Open a new chatbot conversation and explain the activity. Tell the chatbot that the passage contains three deliberate errors, but instruct it not to identify them. Instead, its job is to discuss the passage with the student as the student searches for the mistakes.

Paste in the passage and turn the computer over to the student.

Now the chatbot isn’t doing the homework.

The student is checking its homework.

Why Finding Mistakes Can Help

Learning by identifying errors predates modern AI.

Researchers writing in the journal Computers in Human Behavior studied 208 middle school students during lessons involving decimals.

One group examined work attributed to a fictional student and had to find and explain the errors.

Another group solved equivalent math problems and received feedback on its answers.

The students who searched for and explained mistakes performed better on a delayed test. They were also better at determining whether their own answers were correct.

There was a tradeoff.

Students who simply solved the problems reported enjoying the lessons more.

Error hunting wasn’t as much fun.

But it produced better results on those measures.

That makes the approach particularly interesting for parents and teachers thinking about how AI can fit into education.

A chatbot can generate an essay, solve many homework problems, and answer factual questions in seconds. Used that way, it can sometimes eliminate the very thinking an assignment was designed to require.

An error hunt reverses the roles.

The computer supplies the work.

The student evaluates it.

Make The Mistakes Clear

The activity’s quality depends on the errors.

They should be factual mistakes with clear answers, not matters of opinion or interpretation.

They also need to match the student’s age and what has actually been covered in class.

For a younger student, the errors may need to be obvious enough to identify from a textbook, class notes, or assigned reading. Older students can handle more subtle mistakes that require them to compare dates, causes, definitions, or other details.

Parents may need to experiment with a few passages before finding the right difficulty.

There is also a good reason to keep the answer key.

Chatbots aren’t always reliable when following instructions that require exact counts. Ask for three deliberate errors and a chatbot may occasionally include a fourth mistake without realizing it. It may also make an accidental factual error in addition to the ones it intentionally planted.

Parents should therefore review the passage and answer key before giving the exercise to a child, especially when accuracy matters.

The activity can also be done without allowing the chatbot to guide the student at all. A parent can simply print the flawed passage, ask the student to circle the errors, and explain each correction.

However, the AI conversation adds another possibility.

Instead of immediately telling the student whether an answer is correct, the chatbot can be instructed to ask questions: What makes you think that statement is wrong? What do your notes say? What date should appear there? Can you explain why?

That puts the emphasis on reasoning rather than guessing.

And the student still has to know the material.

Someone who knows nothing about the Missouri Compromise will have a hard time finding believable errors about it. The same is true for photosynthesis, decimals, or the Dust Bowl.

The technology doesn’t eliminate the test.

It disguises it.

The student isn’t being asked to recite everything learned in class. Instead, the challenge is to catch someone — or something — getting it wrong.

That can make AI useful for a homework activity precisely because the student isn’t supposed to trust it.

The assignment becomes: Prove the computer wrong.

And to do that, the student first has to know what’s right.