Thousands of Irish teenagers will soon be sitting down to chew their pencils and write essays for their Leaving and Junior Certs. They won’t be able to access ChatGPT, Claude or Copilot in their school halls.
But because of the increasing amount of home-based project work in the new curriculums, more and more students will be able to use these AI tools in work submitted in advance of the written papers. Apart from the examiner’s nightmare of keeping a fair and level playing field in assessing these projects, what are the implications for the students themselves, in using AI?
Research from Massachusetts Institute of Technology (MIT) using the Scholastic Assessment Test (SAT) has recently thrown some light on this. The SAT is the standard assessment for entry to a US college and is comparable to some Leaving Certificate essays with questions like: “Does true loyalty require unconditional support?” Or, “Should people who are more fortunate than others have more of a moral obligation to help those who are less fortunate?”
It’s a sunny day and confronted with having to tackle one of these essays you think to yourself, why not just ask ChatGPT or Claude or Co-pilot? Great idea – please write me a 400 word essay on this topic. Five seconds later you are reading a beautifully polished, well-argued essay on the question of loyalty or moral obligation of the more fortunate. Whoopee – out into the sun.
READ MORE
What’s the harm in that? Isn’t everyone else doing it, you rationalise. Nobody will have to write this stuff in the real world any more – why would they, when machines can do it so much better?
And you’ve saved yourself a couple of gruelling thinking hours involving false starts, scored-out sentences and stress-inducing mental blocks, haven’t you? Not to mention the vitamin D soaking into your skin – isn’t that supposed to be good for your brain?
Before I answer that, a word about your brain. The 80 billion brain cells are connected with each other by billions of long white “cables” called axons. When you think, imagine, remember, visualise, listen, watch or puzzle, electrochemical signals whizz up and down these long cables, from the back to the front of your brain, from left to right and in every other direction. They are like the M7 to Limerick, only faster.
But while the M7 gets gradually worn out with all that traffic and needs road repairs, the opposite happens to the motorways of the brain: the more traffic there is, the stronger these cables become. Thinking and imagining stimulates little cells called oligodendrocytes to strengthen the insulation – called myelin – on the cables. This makes the signals faster and cleaner so you think and imagine faster and better, the more you do of it.
This is what good education does to your brain – it sculpts it. Teachers are like neurosurgeons, but better at building brains. And this is why good education makes countries healthier, longer-living, happier and richer. Just like Ireland over the last half century.
[ Oracle to shed 150 jobs in Ireland as it grapples with AI cash crunchOpens in new window ]
So, what happens to these brain superhighways when we use large language models? Researchers at MIT’s Media Lab studied students’ brains using EEG (which records the electrical activity of the brain using small, metal electrodes attached to the scalp) as they wrote SAT essays like the ones mentioned above. Some wrote the essays without any technology available, while others had access to ChatGPT. In a final session they switched, with the no-technology group now accessing ChatGPT and the ChatGPT group now having to write without the help of a machine.
The result? Traffic slowed on the brain’s M7. The brain cells talked to each other roughly 40 per cent less when relying on ChatGPT. The hum of the superhighway traffic died down. Students remembered less of what they had written. Even worse, the ones who had used ChatGPT and switched to brain-only still showed less traffic up and down their brain’s cabling. It makes sense if you think about it – who remembers their friends’ phone numbers when they are stored in their phones? Why bother thinking hard when the machine in your pocket can do it for you?
Just as we take exercise to keep our heart and body fit, so our brains need exercise even more to keep their superhighways fast and clean. But every time we take the easy route and let a machine think for us first rather than reviewing our work last, we risk weakening the road networks of our brains.
What’s the harm in taking a shortcut for writing an essay when that’s not something you’re likely to be doing for the rest of your life? Unfortunately, a corrosive mental passivity may extend into other areas. Researchers in Oxford University let students solve arithmetic fraction problems with the help of an AI assistant, but then took the AI away, leaving them to rely on their brains only. The result? Their performance sank and they didn’t persist with difficult problems compared to a no-AI control group. Worse still, in another experiment, relying on AI for arithmetic problems made them worse at a reading comprehension tests.
If offloading our thinking to a machine can lead to us becoming mentally passive in other domains, this has huge implications not just for academic performance, but for our ability to live happy lives. Why? Because life throws up problems that we have to solve, but if we have lost the habit of problem-solving ourselves and instead go first to machines, then we risk harming our emotional lives too.
Does this all mean we should never use artificial intelligence? Of course not, it’s here to stay. What it does mean is that we should always let our brains do the work first and then use AI as an assistant or a collaborator to check and improve our work. The Oxford study found that the harmful effects of AI were largely confined to those who went to it for immediate solutions rather than using it as an assistant after they had worked on the arithmetic fraction problems.
When we use AI as an assistant or collaborator, we get the best of both worlds: we have fed our brain the exercise it needs and we have benefited from the wider perspective of an intelligent machine. Rather than cognitive surrender, we have the possibility of using AI to supercharge our intelligence and our ability to solve problems.
But this requires effort and effort is ... well ... effortful. Many students will take the easy route for their assessments and it will be hard to distinguish their passively generated essays and projects from their peers who have used their brains first and AI second.
But in the long term those who surrender will pay the price as the motorways of their brains slow down. Our teacher-neurosurgeons have their work cut out to stop this happening. But if they do, artificial intelligence offers incredible opportunities to help them extend their students’ intelligence and expand the motorways of their brains.
Ian Robertson is emeritus professor of psychology at Trinity College Dublin and author of How Confidence Works
