Leah is revising for her 'Romeo and Juliet' mock exam. Two hours before bed, with the exam in a fortnight, she starts writing and revising about the theme of fate.
She asks her AI chatbot for the key quotations and what they show. Back comes a tidy list of the classic quotes: "star-crossed lovers", "fortune's fool", and "a plague o' both your houses". Each come ready-made with a line of analysis that is clearer than her own messy notes. She reads them, recognises all three, then rereads them a couple of times so that they stick.
To really make sure she is ready, she gets her chatbot to generate a grade 9 essay, with all the powerful phrases she needs. It's lengthy, but she reads, rereads and highlights.
She feels so much more confident now, so Leah shifts to revising the theme of conflict.
In the mock, she remembers the quotations well enough, and has some ideas, but she can't seem to knit them together into the bigger picture of an essay. Leah cannot explore Shakespeare is doing with them. She recalls fragments from the grade 9 essay she read and reread, but she cannot seem to graft it into the different question she has right now, that isn't quite about fate or conflict.
AI and easy learning
Every generation of teachers has bemoaned the potential damage of using calculators, asking Google, or consulting YouTube explainers. There were no doubt losses with each new glossy tool, but there was also no crisis in education along the way either.
Small clues have alerted us to the issues of offloading thinking to technology. A study aptly titled ‘Information without knowledge’, from 2021, showed that students preparing for quizzes who searched online to help were overconfident and did worse than those who studied the materials directly. It's now a warning from history of the dangers of outsourcing our thinking.
With the power of AI breaking new ground daily, there is a danger that it is poorly used by novice students, and that it replaces the hard-won acquisition of knowledge that is needed to learn.
There is an understandable worry about cognitive offloading, which describes the act of letting AI do the hard thinking for you. In recent research with secondary school age students, they preferred using AI than effortful note-taking, but this was to the detriment of their learning. Additionally, when secondary school students used AI to help with their maths, it wasn't a problem, until that support was taken away, and then they performed worse than those who didn't use AI.
If we wrongly assume that easy learning is better, then students are likely to offload the hardest topics and tasks to AI and infer that they are better off because of it.
The problem is that the opposite is true: the harder it feels the better, if we experience success with the task.
Let's return to Leah and her studying Shakespeare. If 'fate' is her trickiest topic, she has done a good job monitoring, and she is exercising what is fancily termed as metacognitive regulation (monitoring your own thinking while you work). But then if you let AI do the thinking for the hard stuff, short term successes are traded for the potential of long-term failures.
Ultimately, we need to apply the Goldilock's principle to independent learning. Too easy, and we don't learn much. Too hard, and we fail, and don't learn much either. We need some 'deliberate difficulty'. Just enough to make us think hard and grapple with the information at hand, but not too much so that it sees us fail.
Ample amounts of cognitive science research has shown the potential power of retrieval practice, spacing, self-explanation, and more. These are all strategies that make learning harder in the short term. Of course, for a tired teenager, and even a time-poor adult, it makes sense that we struggle with these strategies when working independently.
Alas, AI may make learning all too easy.
Puncturing overconfidence and using AI effectively
Problematically, there is research to show that people are most likely to offload difficult learning when they doubt their own performance. If even experienced adults are likely to pass over over tasks they are perfectly able to do to AI, because those tasks feel hard, why would we expect a student working in their precious independent time to act differently?
Simply being told not to use AI is unlikely to work. We know how that can backfire with teens for a host of their habits. Instead, we need to be explicit about how to learn independently and how to use AI effectively.
Handy research on dealing with reading misinformation and exercising metacognitive regulation is useful here. Research from Salovich and Rapp found that warnings to readers about falsehoods in their reading usually fails. However, they found that if you posed them reflection questions before reading (such as 'when did you last read something inaccurate? When did you last rely on it, why did you rely on it, what did you learn? etc.) they made half as many errors in their reading and their confidence was helpfully recalibrated too.
Similarly, when using AI help, students could be trained to use prompts that elicit questions and reflection, rather than simply posing them the right answer or the grade 9 essay.
Independent learning: avoiding offloading
The simple question is: does the hard thinking still happen in the Leah's head, or is it being enacted by the AI chatbot?
We can encourage Leah to be more metacognitive, monitoring and reflecting on her revision to better effect.
For example, Leah asks for five key quotations on fate, but before reading a word of analysis. She then types her own response to each, from memory, in her own words. Then she sense-check against the tool. Thinking comes first, with feedback or the 'answer' coming second. The effortful retrieval survives, and she finds out immediately where she has gaps or struggled.
Downloading a grade 9 essay and rereading it unlikely to be effective. Dunlosky and colleagues rate rereading among the least effective strategies we have for long-term learning. Leah should instead look for question prompts on the essay title, and write an answer, before typing it up and improving it along the way. Then she might test it with a chatbot for feedback against the mark scheme (a chatbot could help here if a teacher is unavailable - the latter being preferable for obvious reasons).
Both examples start with Leah working and thinking. Independent work is scaffolded but not replacement by a crutch.
Pupils will not naturally defer to these strategies. They are 'cognitive misers' who want to save energy where they can. It has to be taught and wedded closely to their subject specific tasks. It needs modelling and scaffolding too for it to be more likely to stick as a routine habit.
Is AI going to kill off independent study? The answer is probably no. But there is a lot of work for teachers, and their students, to do if they are to use AI productively, and not see it quietly damage independent learning.
Related reading:
- I've written about the importance of 'Learning by Hand' and how using tech may miss vital learning developments - see HERE.
- My book, 'Why Learning Fails (And What To Do About It)', includes chapters on tackling independent learning and student overconfidence – which AI can feed if we are not careful – see HERE.
Comments