A real and growing problem is being widely discussed in technical education circles right now: students who lean heavily on AI assistants like ChatGPT, Claude, and Copilot are failing exams at higher rates and showing measurably weaker math and problem-solving skills. Reports from Berkeley computer science classes have brought the issue into sharp focus, but the pattern is showing up across universities everywhere. If you are a student noticing your grades slipping despite feeling productive with AI tools, or you are watching your foundational math intuition erode, this guide walks through why it is happening and what to do about it.
This is not a moral panic. It is a measurable shift in how learning is being short-circuited, and the fix is technical, behavioural, and practical. Below is a structured troubleshooting approach that treats the problem the same way you would diagnose a broken build pipeline.
What Causes This Issue
The core problem is cognitive offloading. When you ask an AI to solve a derivative, debug a recursion, or write a proof, your brain skips the exact struggle phase that builds long-term retention. The friction is the feature. Remove the friction and you remove the learning.
Several specific failure modes are showing up repeatedly in classrooms:
- Illusion of competence: Reading AI-generated solutions feels like understanding, but recognition is not recall. On a closed-book exam, the scaffolding disappears.
- Skill atrophy in prerequisites: Calculus, linear algebra, and discrete math intuitions weaken when every integral or proof is delegated. By the time upper-division algorithms or machine learning courses arrive, the foundation is gone.
- Pattern-matching without modeling: Students learn to prompt rather than to reason. They can describe a problem to an AI but cannot construct a mental model of it themselves.
- Hallucination acceptance: AI outputs contain subtle errors in math derivations, off-by-one bugs, and incorrect complexity analyses. Without strong fundamentals, students cannot catch these, so wrong answers get submitted and wrong intuitions get internalised.
- Time displacement: Hours that used to go into wrestling with problem sets now go into prompt engineering and copy-paste cycles, which feel productive but produce no durable skill.
Users in the Apple Support Community discussing this trend have pointed out an additional factor: on devices like the iPad and Mac, AI assistants are now one tap away through Apple Intelligence, ChatGPT integration, and writing tools. The friction to ask is near zero, which makes self-discipline the only remaining safeguard.
Step-by-Step Fixes
- Audit your current AI usage for one week. Keep a simple log on your iPhone or Mac noting every time you queried an AI for coursework. Categorise each query as either “unblocked me on something I genuinely could not figure out” or “saved me time I should have spent thinking.” Most students discover the ratio is alarming.
- Establish an AI-free first pass on every assignment. Solve the problem on paper or in a plain text editor with no AI tools open. Set a minimum time, typically 25 to 45 minutes per problem, before any external lookup is allowed. Use the Focus mode on macOS or iOS to silence notifications and block AI apps during this window.
- Use Screen Time restrictions on AI apps. On iPhone, iPad, and Mac, open Settings, go to Screen Time, choose App Limits, and add ChatGPT, Claude, Copilot, and browser-based AI sites to a daily limit. Even a 30-minute cap forces intentional use.
- Disable Apple Intelligence writing tools during study sessions. Go to Settings, Apple Intelligence and Siri, and toggle off writing tools temporarily while working through problem sets. The same option exists in System Settings on macOS Sequoia and later.
- Switch AI from solver to tutor. When you do use AI, never ask for the answer. Ask it to ask you Socratic questions, to point out which concept your reasoning is missing, or to grade your written attempt. Prompts like “Do not give me the solution. Ask me one question that helps me see what I am missing” reshape the interaction.
- Rebuild the prerequisites you skipped. If you are in an algorithms class but your discrete math feels shaky, go back to the textbook chapters and work problems by hand. Yes, this is slow. That is the point.
- Practice closed-book under exam conditions weekly. Simulate the test environment. No devices, no notes, timer running. This is the only honest signal of what you actually know.
Additional Solutions
Beyond the immediate behavioural fixes, several structural changes help.
Form a small study group that explicitly bans AI during sessions. Explaining a concept to a peer is one of the highest-retention activities known in education research, and it cannot be outsourced. Use FaceTime or in-person sessions where screens are visible to each other.
Adopt the Feynman technique. After working a problem, write an explanation as if teaching it to someone with no background. If you cannot, you do not understand it, and the gap is exactly where AI was doing the lifting for you.
Use spaced repetition tools like Anki for definitions, theorems, and core algorithms. Active recall on a schedule beats re-reading or re-prompting an AI every time.
For coding specifically, turn off inline AI autocomplete in your editor. Copilot, Cursor, and similar tools generate plausible code faster than you can think about whether it is correct. In Xcode, disable predictive code completion under settings during learning sessions. Write code from scratch, then compare to AI suggestions afterward as a review exercise.
Track your work on paper. Physical notebooks for math derivations and algorithm sketches force a slower, more deliberate process that screens do not. The Apple Pencil on iPad with a notes app can approximate this if paper is impractical.
Finally, talk to your instructor. Many professors are restructuring courses around the assumption that AI is available, which often means heavier weight on in-class exams, oral defences of submitted work, and proctored assessments. Knowing the grading reality helps you calibrate effort.
When to Contact Apple Support
If Screen Time limits or Apple Intelligence toggles are not behaving as expected, for example, App Limits not enforcing on AI applications, Focus modes not blocking specified apps, or Apple Intelligence writing tools reappearing after being disabled, this is a device-level issue worth escalating. Contact Apple Support through the Apple Support app on your iPhone, by calling directly, or via the official support website. Provide your device model, iOS or macOS version, and screenshots of the misbehaving settings. For managed school devices, your institution’s IT department may have MDM policies overriding personal Screen Time settings, and they should be your first contact.
FAQ
Is using AI for coursework always bad? No. The problem is using it as a replacement for thinking, not as a supplement. After you have struggled, AI is excellent for clarifying a stuck point or explaining an alternative approach.
How do I know if I am over-relying on AI? The clearest test is closed-book performance. If your homework grades are strong but your exam grades are collapsing, the gap is your answer.
Will professors detect AI use? Detection is unreliable, but that is the wrong framing. The exam is the detector. Skills you did not build will surface there regardless of policy.
Can I use AI to study for exams? Yes, as a quiz generator or tutor that asks you questions. Not as an answer key.
What if my whole class uses AI and I fall behind by not using it? You will likely match their homework grades with more effort, then significantly outperform them on exams and in later courses. The compounding favours the foundation builders.
