AI Completes My Homework — But Is It Stealing My Brain? The Learning-Performance Paradox

Forum topic · QianXun · 2026-05-08

Summary

A Chinese tech forum post discusses the learning-performance paradox raised in the 2026 paper "Building AI Companions that Prioritise Learning over Performance" by Hassan Khosravi. Using a GPS-versus-paper-map analogy, the author argues that performance-oriented AI (like current LLMs that instantly deliver answers) maximizes task completion but undermines genuine learning. The post connects this to the psychology concept of "desirable difficulties" — real learning requires effortful thinking and struggle — and warns that overly helpful AI fosters "metacognitive laziness." The proposed solution is shifting from task-oriented AI to AI Learning Companions (ALC) that use Socratic questioning, retrieval practice, self-explanation, and guided reflection instead of handing over answers. The author cautions against a future where AI writes student assignments and teachers grade with AI, leaving knowledge circulating between models while humans learn nothing, and concludes that in an era of instant answers, protecting our right to think slowly is essential.

Overview

A recent zhichai.net forum post explores a thought-provoking question: AI helps us finish homework effortlessly, but does it quietly erode our ability to learn? The post is built around the 2026 paper "Building AI Companions that Prioritise Learning over Performance" by Hassan Khosravi.

The GPS vs. Paper Map Analogy

The author opens with a travel scenario:

1. Full GPS navigation: fast, stress-free, and you arrive precisely — but you remain unable to find your way without the screen. 2. A paper map: slow, confusing, with wrong turns — but after one trip, the city's layout is etched into your memory.

Performance (reaching the destination): GPS wins.
Learning (knowing the city): the paper map wins.

This captures the paper's central idea: the learning-performance paradox.

What Is the Learning-Performance Paradox?

Author Hassan Khosravi observes that almost all current LLM usage is performance-oriented:

Need an essay? AI writes it instantly.
Stuck on a math problem? AI outputs the steps.
Buggy code? AI fixes it immediately.

Your tasks get done beautifully — but did you actually learn? Probably not.

The post links this to the psychological concept of "Desirable Difficulties": genuine learning requires painful thinking, repeated attempts, and the mental strain of being *almost* but not quite understanding something. That struggle is the ticket to long-term memory. Today's AI is so eager to help that it dissolves all the difficulty, making our brains "metacognitively lazy."

From Tool AI to AI Learning Companion (ALC)

The paper argues we should stop treating AI as an answer generator and turn it into an AI Learning Companion (ALC). The contrast:

Task-oriented AI (today's GPT):
You: "How do I solve this?"
AI: "The answer is X, here are steps 1-2-3."
(You: Ctrl+C, Ctrl+V — brain offline.)
Learning-companion AI (the future ALC):
You: "How do I solve this?"
AI: "Look at this formula. Remember the principle we covered yesterday? What happens if you substitute A into B?"
You: "Hmm, I think it becomes C."
AI: "Nice! What would you do next with D?"
(You: forced to think — brain online.)

Such an AI deliberately "makes things hard" for you, guiding you toward answers through retrieval practice, self-explanation, and collaborative insight.

Why It Matters for Everyone

If future education becomes "AI writes student homework, teachers grade with AI," knowledge merely flows between two models — human students become hollowed-out containers. The post quotes Feynman: people don't learn through understanding, but through memorizing or some strange ritual.

The paper's reminder: in an age when AI can produce instant answers, what we most need to protect is the right to think slowly and figure things out gradually.

The Takeaway

> Good tools extend your hands and feet; mediocre tools shrink your brain.

Before asking AI for the answer, ask yourself: am I looking for a GPS that gets me to the destination, or a guide who teaches me to read the map?

Don't let AI finish all our work — and steal our growth along the way.