AI Detectors Are as Morally Wrong as the Cheating Done by Students Who Submit AI-Generated Work as Their Own

For me, using AI detectors to determine whether a student forged an assignment using Generative AI tools is morally wrong. 

I agree with Carissa Veliz, who writes in her book “Prophecy: Prediction, Power, and the Fight for the Future, from Ancient Oracles to AI”:

“A predictive approach to ethics is likewise inadequate for matters of justice, inside and outside the courtroom. In criminal contexts, merely statistical evidence isn’t enough.”

Because AI detectors use statistics and probability to predict whether a student’s work is AI generated or not, it alone should never be used as the sole evidence for making the determination on whether the student cheated and turned in AI work as their own. This probability of having AI generated material has room for error, and when it comes to dispensing justice, it is inadequate for me. I would not use it alone for detecting whether a student engaged in this unethical behavior. 

In some ways it would seem to me to be akin to the lie detector, which attempts to detect patterns of truthfulness and untruthfulness but can’t tell you whether a person is being deceptive about a specific instance.

Do we then just accept the student’s work as is? Unless we can find some causal evidence, not probable evidence, I think we have little choice, but we can devise ways to ask the student to defend their work and ensure they have invested their experience fully in the learning.

Still, to me the greater problem is that the student chose to cheat to begin with. It is a moral and a trust issue. It is a symptom of a character concern in that student that they would resort to such action, and from a societal standpoint, that should be of equal concern, that a student would choose that course of action to begin with.

What seems like a better course of action rather than simply accusing the student of cheating based on a technology, would be to devise a way that the student must defend their work, without assistance. 

For example, it could be a panel of teachers asking questions designed to ensure that the student was knowledgeable about their work. Criteria could be determined ahead of time that outlines what a successful defense of the work looks like, and the final assessment on the student’s work would be based on that alone.

Ultimately though, we still should acknowledge the moral problem underlying this, which is the same problem that has been beneath cheating since students have been subjected to instruction, which is that a student would deceptively choose to cheat to begin with. AI cheating is in some ways just another high tech cheating tactic.

The solution in this case, is not more technology, though educators, being the tech-solutionists they are, always seem to turn to tech for answers. 

Tech companies love it, because they can sell us a tech that causes a problem, then turn around and sell us another technology to solve that problem, and then another tech to solve that tech’s problem and so on. 

AI detectors are not the answer.

Instead, the answer lies in working toward the goal of helping students become ethically averse to cheating to begin with through moral instruction and character development, educating them to be better than that.

Also, the answer lies in making sure the teaching and learning experience requires the student to demonstrate their learning in ways that can’t be fabricated through AI. This is not a technology problem, but an old educational problem of, “How do I ensure that students have learned?”