A hidden prompt caught a class handing in essays they never read, and showed the real divide in how people use AI.
Some cunning lecturers have started hiding instructions in their assignment questions. The trick is simple: a line of white text on a white background, invisible to the student but carried along when the question is copied and pasted into a chatbot. A professor at Alcorn State University used it in a midterm on the Industrial Revolution. The hidden line told any chatbot reading the prompt to work the word “Madagascar” into its answer, however little sense it made. Thirty-two of the 35 students handed in essays that mentioned Madagascar.
Think about what that means. Madagascar has nothing to do with the Industrial Revolution, and a stray reference to it would leap off the page at anyone who read the essay through once. At least 32 students submitted work they had never read.
It reminded me of the stories that surface every so often of drivers who follow their satnav straight into a lake. When one came up in conversation, a relative of mine quipped that satnav is fine as long as you know where you’re going. Many a true word is spoken in jest.
Back in the 1990s, when the web first arrived, people in the media business drew a line between “lean-back” and “lean-forward” technology. Television was lean-back: you sat on the sofa and it came to you. The desktop computer was lean-forward: you sat up, paid attention and took part, because the information flowed both ways. There was plenty of debate about whether the two could be combined, and how much effort people would put in to be entertained. Experiments such as WebTV came and went. In the end the smartphone and the smart TV settled the question in the living room. We now watch television with a phone in our hand, half-attending to both. Screenwriters have reportedly been asked to have characters spell out what is happening, for the benefit of viewers looking at their phones.
Our attention is split, and I suspect the amount of real thinking we do about what we consume is shrinking with it. I’d wager a fair few of those 32 students had a video playing on the other screen while the chatbot did their homework.
They treated AI as a lean-back technology: something that does the work for you. Type a request, take whatever comes out, hand it in. As long as something essay-shaped emerges from the other end of the sausage machine, the job is done.
Those of us who use AI as a lean-forward technology work differently. We prompt, read the output, rethink the prompt and go again. The output is a draft to argue with, and the arguing is where the thinking happens. The part of us that leans forward is the brain.
To my mind, this is where the real argument about AI sits. I regularly read predictions that AI will be the end of education, or even that humans will evolve smaller heads because we will no longer need our brains. (I do spend time in some odd corners of the internet.) Those predictions assume everyone will lean back. The Madagascar essays show what that looks like. They don’t show that it is inevitable.
It is worth asking why the students leaned back. Our education system was built in the industrial age to prepare young people for the office and the factory. It rewards producing the required piece of work on the required day. Seen that way, the 32 students did exactly what the system taught them: they delivered an essay-shaped object to a deadline. AI simply made it effortless.
The workplace those students are heading into looks very different. It is hard to say what tomorrow’s jobs will be. A job for life has become rare, adaptability is prized, and lifelong learning is less a choice than a condition of earning a living. Responsibility for staying employable has shifted too. It used to be shared by the state, through education, and by employers, through apprenticeships and training. Increasingly it lands on the individual.
If learning has become an individual responsibility, it makes little sense to keep teaching as though every child is on the same page. In some countries that is literal: a whole year group sits down at the same time and opens the same textbook at the same page.
This is where AI could be transformative. Used the lean-forward way, it is a personal tutor that responds question by question, minute by minute, in a way no teacher with a class of 30 ever could. For that to work, students would need to take charge of their own learning from an early age, with schools organised to support them. How we get there deserves a post of its own.
I’ll declare my bias. Writing with AI has changed how I work and how I think. I believe it could do the same for most people, provided we keep alive the curiosity children start out with, because curiosity is what makes you lean forward.
Satnav is fine as long as you know where you’re going. So is AI.