Do Kids Still Need to Do Homework in the AI Era? Educational Reflections from a Lex Fridman Conversation

AI is reshaping the underlying logic of education. When large models can break down a math paper into personalized explanations within a learner's "zone of proximal development" at any moment, and when calculators have long been able to perform all four arithmetic operations, why do we still make children practice repeatedly? This conversation from the Lex Fridman podcast offers a thought-provoking answer: the more powerful the tool, the more people need to retain the ability to "do it themselves."

From "Can't Understand the First Sentence of a Paper" to Personalized Learning

The guest mentions a real scenario: a graduate student gets a paper and a week later is still stuck on the first paragraph—every term in the sentence "perturbation theory for nonlocal systems" is unfamiliar, so they grind to a halt. The traditional way to read mathematics is to skip the intricate opening, jump straight to Lemma 3, and work backward from the part you can actually manipulate to deduce what the whole argument is doing.

LLMs change this process. Throw the paper at the model, start asking questions, and it quickly locates the learner's "zone of proximal development"—that stretch between what you already know and what you should learn next. This capability isn't just for kids; adults changing careers or learning new fields benefit just as much. Previously, a graduate student had to "completely rewire their brain" to enter a new area of mathematics; now that rewiring process can be significantly accelerated.

Another learning path comes from the conversation's preference for visualization: build the simplest toy example, ideally one you can interact with and drag parameters on. Abstract symbols and notation established by community convention are often the barrier for beginners—once you know "what this little symbol stands for," the entire knowledge system quickly falls into place. This kind of play gives people the confidence that "I can tackle any field"—even something completely unfamiliar like protein folding.

Zone of Proximal Development (ZPD) was proposed by Soviet psychologist Vygotsky and refers to the gap between a learner's actual level of independent problem-solving and their potential level under guidance. Teaching that falls within this zone is most efficient; below it is boring, above it is frustrating. The article borrows this concept to describe the unique value of LLMs: the model can quickly diagnose a learner's current knowledge boundary through dialogue and slice material into a granularity that is "a stretch but reachable," rather than throwing a complete paper at them all at once. This also explains why the same paper can deter one person at the first paragraph while another can dismantle it layer by layer with guidance.

The Real Bottleneck Isn't the Tool, It's Motivation

Whether a tool can accelerate learning depends on whether the person is willing to learn. The conversation points directly at the "human motivation problem."

Three key capabilities are named: curiosity, asking good questions, and the critical ability to "tell bullshit when you see it." This is called "consulting skills"—acquiring information in a field where you are not an expert and judging whether to dig deeper or push back.

This is precisely the hardest part of education. When undergraduates or even middle schoolers use these tools to learn, their brains aren't fully developed, yet they're expected to have "consulting skills" beyond critical reasoning. How to teach children to judge whether information is true or false, and how to question it, is a massive task facing the education system.

The conversation even suggests that the hardest problem isn't in undergraduate education but in middle school: when does homework start getting hard? At that point, if AI can help with everything, why should kids still do homework?

Consulting skills in this context are not business consulting techniques but the ability to efficiently acquire and evaluate information in a non-expert field: knowing what to ask, how to follow up, and how to identify gaps and hallucinations in answers. It complements traditional "subject knowledge"—the former determines whether you can enter an unfamiliar field, the latter determines how deep you can go once you're in. Large models lower the barrier to acquiring information but also amplify the risk of credulity, making the judgment of "whether to dig deeper or push back" a core competency. The article emphasizes that this skill needs to be cultivated in childhood because it relies on long-term training in critical reasoning, not on cramming as an adult.

The Seventy-Year Lesson of the Calculator: Skills You Skip Become Costs

Mathematics "started earliest" in this discussion—calculators have been around for about seventy years, yet many schools still require children to memorize multiplication tables.

The guest's attitude is clear: those voices arguing that "with calculators, there's no need to memorize multiplication tables" have made the children who listened pay a real price, causing "a great deal of human suffering." His principle: just because a calculator can do it doesn't mean you don't need to be able to do it. In actual calculation you can of course use tools, but in principle you should have the ability to work it out yourself.

This differs from the argument that "only forced suffering has value." The guest admits that he himself only appreciated the value of difficult tasks because he was forced into them, but he emphasizes another fact even more: children are born loving to master skills. The excitement of tying your shoes for the first time seems trivial to adults but is "the greatest thing" to a child. Every step toward independence is worth celebrating—and independence from machines means that just because a machine can do something doesn't mean you don't need to retain the ability to do it yourself.

The multiplication table vs. calculator debate is one of the longest-running cases in the history of educational technology. Supporters argue that mechanical calculation can be outsourced and students should invest time in conceptual understanding; opponents point out that mental arithmetic is the underlying support for number sense, estimation, and algebraic operations, and skipping it weakens later learning. The position cited in the article is closer to the latter: tools can be used for actual calculation, but the ability to "work it out yourself" should be retained as a principle. This argument can be extended to writing, programming, and other fields—tools replace the execution step, not the understanding step; once understanding is outsourced, learners lose the ability to judge whether the tool's output is correct.

Implications for AI Education

This conversation doesn't give a simple answer to "should homework be abolished," but it provides a framework for judgment: AI as an accelerator of the "zone of proximal development" is valuable only insofar as the user has curiosity, questioning ability, and critical judgment. The real task of the education system is to start training these "consulting skills" while children are still young, and at the same time retain the fundamentals that seem replaceable by tools but actually constitute the underlying capabilities of thought. The stronger the tool, the more we must guard against outsourcing "being able to do it."

One-line Summary

The more powerful the tool, the more people need to retain the ability to do it themselves.

Paragraph Summary

Based on a Lex Fridman podcast conversation, the article explores how AI is reshaping education. LLMs can break papers down to a learner's "zone of proximal development," significantly accelerating the learning of hard subjects like mathematics through questioning and visual toy examples. But the real bottleneck isn't the tool—it's human motivation: curiosity, the ability to ask good questions, and the "consulting skills" to judge whether information is true or false. The seventy-year debate over calculators and multiplication tables shows that abandoning skills tools can perform comes at a real cost. The core principle: just because a machine can do something doesn't mean a person doesn't need to retain the ability to do it independently; the education system should train consulting skills and fundamentals from an early age.