How AI is changing education and study

Before you start

Year band: Years 10–12
Estimated study time: 25 minutes
Prerequisite topics: You should understand the difference between passive review (such as re-reading notes, highlighting text, or reading AI-generated summaries) and active retrieval (testing yourself, writing from memory, and solving problems unassisted). You should also be familiar with the forgetting curve and the concept of "desirable difficulties"—the idea that mental effort during study directly correlates with long-term memory consolidation.
What you will need: A notebook, a pen, and a willingness to embrace productive struggle.

What you'll learn

• Explain how unrestricted AI use acts as a cognitive crutch and degrades long-term retention.
• Identify the difference between cognitive offloading and active retrieval.
• Apply Socratic prompting strategies to turn AI into a feedback tool rather than an answer generator.
• Design a independent, 30-minute AI study block to prepare for closed-book exams.

Core concepts

The Illusion of Competence and Cognitive Offloading

As artificial intelligence tools become integrated into our daily study routines, many students ask: how will AI change education and the way people learn? The answer lies in how we handle the effort of learning. Unrestricted AI chat often raises short-term assisted performance, but it can severely harm long-term, unassisted learning. When you ask an AI to write an essay or solve a math problem, you engage in cognitive offloading. This process bypasses the mental strain required to build durable memory pathways.

Recent cognitive science research highlights this danger. In a landmark study by Bastani et al. (PNAS 2025), high school math students who used unrestricted GPT-4 during practice scored 17% worse on unassisted exams than those who studied without AI. Similarly, a study by Barcaui (SSHO 2025) revealed an 11% percentage-point retention deficit on a surprise 45-day test. Traditional learners scored 68.5%, while unrestricted ChatGPT users scored only 57.5%. Even though the AI-assisted students reduced their study time by 45%, this efficiency failed to protect them from long-term retention deficits.

Metacognitive Laziness and the Safety Gap

When AI does the heavy lifting, students often fall victim to what Fan et al. (BJET 2025) call "metacognitive laziness." This occurs when learners stop monitoring and evaluating their own thinking because they trust the AI's fluent responses. This creates an epistemic discrepancy known as the "Safety Gap" (Wang & Shan, Frontiers in Education 2026). The Safety Gap is the difference between your polished, AI-assisted homework performance and your actual, internal ability to verify and reproduce that work under exam conditions.

This disconnect has real-world consequences. A large-scale 2026 longitudinal study of 26,811 secondary students found a 20% drop in monthly exam scores within six months of adopting AI for homework, despite their homework scores rising by 18%. Because the AI-generated work looked perfect, students felt a false sense of security—an illusion of competence—only to blank out during closed-book school assessments.

Desirable Difficulties and Active Retrieval

To learn deeply, we must embrace what psychologists call "desirable difficulties." This principle states that the active effort required to retrieve information from memory actually strengthens neural connections. Reading an AI-generated summary is passive; retrieving the facts from your own brain is active. The ACARA National Curriculum General Capabilities emphasize critical and creative thinking, which requires students to independently analyze, synthesize, and evaluate information rather than merely consuming pre-packaged answers.

Independent AI Study Rules

To prevent AI from degrading your memory, you must establish a independent study workflow. Rule 1: Never ask for the answer first. Always prompt the AI to withhold solutions and only provide sequential, conceptual hints. Rule 2: Separate retrieval from verification. Write your initial essay outline or math attempt completely closed-book before pasting it into an AI tool. Rule 3: Build a physical Mistake Book. Write down AI-generated corrections by hand to force deep cognitive processing. Rule 4: Audit the AI's output. Actively look for hallucinations, over-simplifications, and sycophancy (the AI agreeing with your incorrect assumptions) using your official syllabus and textbooks.

Worked examples

Example 1: Year 12 English Analytical Essay Preparation

A student is preparing for a Year 12 English exam on Shakespeare's Macbeth. Instead of asking an AI to "write an essay on Macbeth," they use a independent, retrieval-first approach.

1. The student closes all books and writes a 3-paragraph essay outline from memory, including a thesis statement, three main arguments, and two quotes they recall.
2. They open the AI tool and input the following prompt: "Critique my thesis statement and suggest three historical context links without rewriting my text. Do not write the essay for me."
3. The student reviews the AI's feedback, identifies gaps in their contextual knowledge, and manually updates their physical study notes.

Example 2: Socratic Physics Problem Solving

A student struggling with a complex physics problem involving centripetal force wants to understand the derivation rather than just copying the formula.

1. The student prompts the AI: "Act as a Socratic physics tutor. Do not give me the formula or the answer. Ask me guiding questions one at a time to help me derive the formula for centripetal force based on Newton's second law."
2. The AI asks: "What is Newton's second law of motion, and how do we define acceleration in uniform circular motion?"
3. The student writes down the answers (F = ma, and a = v^2 / r) from memory, inputs them, and the AI guides them to substitute the acceleration into the force equation to derive F = mv^2 / r. This active struggle ensures the derivation is consolidated in long-term memory.

In tests and exams

School assessments are designed to test your unassisted cognitive ability under strict, closed-book conditions. Here is how passive, AI-assisted preparation fails during exams:

• Closed-book essay writing: Examiners reward deep structural planning, unique voice, and precise textual analysis. If you rely on AI to write your drafts, you bypass the planning phase, leading to "metacognitive laziness" and an inability to structure an essay under timed conditions.
• Multi-step problem-solving: In mathematics and sciences, examiners award marks for intermediate steps and logical reasoning. If you use AI to solve homework problems instantly, you fail to build the necessary schema, leading to a complete block during exams.
• Thinking skills and logical reasoning: High-stakes tests reward independent abstract reasoning. You can browse every supported practice test type to see how these unassisted reasoning skills are assessed.

Practice

1. Warm-up: Read this student's study diary entry: "Today I used ChatGPT to summarize three chapters of my history textbook. I highlighted the key points on my screen and read through them twice." Identify the passive study habits in this entry and explain why they will not lead to long-term retention.
2. Warm-up: Rewrite this passive prompt into an active, Socratic prompt: "Explain the process of photosynthesis and give me the chemical equation."
3. Standard: Design a 30-minute independent AI study block to learn the concept of chemical equilibrium in Chemistry. Break down the time allocation and specify what is done closed-book versus with the AI.
4. Standard: A student gets 95% on their take-home math assignments by using AI to check every step as they work. However, they score 55% on their in-class exam. Diagnose this student's learning issue using the concepts of the "Safety Gap" and the "Illusion of Competence."
5. Challenge: Create an active-learning plan using AI to prepare for an unseen text analysis exam in English. Your plan must include a step where you get AI writing feedback on your work without allowing the AI to write any part of the analysis for you.

Answers

1. Warm-up Answer: The passive study habits include: (1) Offloading the summarization process to ChatGPT, which deprives the brain of the effort required to synthesize information. (2) Highlighting text on a screen, which is a low-utility passive strategy. (3) Re-reading the summaries, which creates an illusion of competence without testing actual recall. To fix this, the student should read the textbook chapters, close the book, and write down a summary from memory before using AI to check for missing details.
2. Warm-up Answer: An active, Socratic prompt would be: "Act as a Socratic biology tutor. Do not explain photosynthesis or give me the chemical equation. Instead, ask me three sequential questions to test my current understanding of the light-dependent and light-independent reactions, and guide me to write the balanced chemical equation myself."
3. Standard Answer: A highly effective 30-minute independent AI study block: (1) 0–15 minutes (Closed-book active recall): Write down everything you know about Le Chatelier's principle and chemical equilibrium on a blank sheet of paper. Attempt to write the equilibrium constant expression (Kc) from memory. (2) 15–25 minutes (Socratic AI feedback): Paste your written recall into the AI. Use the prompt: "Analyze my explanation of Le Chatelier's principle. Identify any logical gaps or missing terms without rewriting it for me. Ask me one guiding question to help me correct my errors." (3) 25–30 minutes (Spaced review): Close the laptop. Handwrite the corrected concepts and formulas in a physical Mistake Book spaced review to consolidate the neural pathways.
4. Standard Answer: This student is suffering from the "Illusion of Competence" caused by "cognitive offloading." By using AI to verify every step during homework, they never experienced the "productive struggle" required to build independent problem-solving schemas. This created a massive "Safety Gap"—their assisted performance was high (95%), but their unassisted performance was low (55%) because they lacked the internal ability to verify their own steps under closed-book exam conditions.
5. Challenge Answer: A comprehensive active-learning plan for unseen text analysis: (1) Step 1 (Unassisted Attempt): Spend 15 minutes reading an unseen poem or prose extract. Write a 200-word analysis focusing on how the author uses metaphor and tone to convey theme, completely unassisted and under timed conditions. (2) Step 2 (Targeted AI Feedback): Paste the unseen text and your analysis into the AI. Prompt: "Act as an experienced English examiner. Grade my analysis based on structural coherence and evidence integration. Do not rewrite my work. Point out two specific areas where my analysis of metaphor could be deeper, and ask me a question to help me expand on them." (3) Step 3 (Manual Revision): Based on the AI's feedback, rewrite the weak sections of your analysis by hand in your notebook, ensuring you actively retrieve the literary terms and structural patterns.

Common mistakes

• Using AI to write the first draft of an essay: This bypasses the critical planning and structuring phase of writing, leading to "metacognitive laziness" and an inability to write under exam conditions. Fix: Plan and write your first draft completely closed-book, then use AI only for targeted feedback on structure and style.
• Asking AI for the immediate solution to a hard math problem: This eliminates the "productive struggle" and prevents the brain from building deep mathematical schemas. Fix: Prompt the AI to act as a Socratic tutor, giving only sequential, conceptual hints to guide you.
• Assuming AI-generated explanations are 100% accurate: This falls into the "verification gap" where you lack the deep knowledge to spot subtle hallucinations or sycophantic errors. Fix: Cross-reference AI outputs with official textbooks, syllabus documents, or reliable curriculum resources.

Quick recap

• Unrestricted AI use acts as a cognitive crutch, reducing unassisted exam scores by up to 17% (Bastani et al., PNAS 2025).
• True learning requires "desirable difficulties"—active retrieval is far superior to passive reading or copying AI-generated summaries.
• Avoid "metacognitive laziness" by always attempting problems unassisted before seeking AI feedback.
• Use Socratic prompts to force the AI to give hints rather than direct answers.
• Maintain a physical Mistake Book to consolidate corrections and bridge the safety gap between assisted and unassisted performance.

To ensure you can perform under strict, closed-book exam conditions without relying on cognitive crutches, you can use RubricMark to test your unassisted skills. Start timed practice mocks and section drills.