The cultural narrative surrounding artificial intelligence in education has reached a fever pitch. Venture capitalists, tech evangelists, and school district administrators frequently speak of Large Language Models (LLMs) as if they are omniscient digital tutors poised to completely restructure, run, and revolutionize the modern classroom. We are told that personalized algorithms will soon replace the messy, intuitive, human art of teaching.

But this sweeping narrative is built on a profound misunderstanding of what these systems actually are. To look at the current state of classroom AI integration is to witness a textbook case of irrational exuberance—a speculative frenzy detached from underlying reality, startlingly reminiscent of the infamous Tulip Mania that gripped 17th-century Holland.

To understand why handing the keys of the classroom over to LLMs is an educational catastrophe in the making, we have to look no further than the insights of neuroscientist Jeff Hawkins and his pioneering Thousand Brains Theory of Intelligence.

The Hawkins Critique: Statistical Mimicry vs. True Intelligence

In a landscape dominated by AI hype, Jeff Hawkins offers a bracing dose of scientific reality. Hawkins argues that despite their fluent prose and seemingly coherent answers, current deep learning models and LLMs are fundamentally not intelligent. Instead, he categorizes them as incredibly sophisticated, brute-force "fancy calculators."

The human brain achieves intelligence by building a dynamic, structural "model of the world" inside the neocortex. This model relies on reference frames and sensory-motor inference—learning through continuous physical movement, exploration, and spatial tracking. The biology is the computation, operating seamlessly on a mere 20 watts of power.

LLMs do none of this. They don't have a model of a student, a model of gravity, or a model of meaning. They are massive, energy-hungry statistical engines designed to predict the next most probable word in a sequence based on vast oceans of static training data.

The Core Disconnect: A calculator mimics the output of mathematical thinking without actually "knowing" what a number represents. Similarly, an LLM mimics the structure of human language without an ounce of genuine comprehension regarding the physical reality behind the words.

The EdTech Bubble and the Ghost of 1637 Holland

When school districts rush to automate lesson planning, implement AI-driven grading rubrics, or introduce virtual AI avatars to lead instruction, they are falling into a trap of speculative value.

In 1630s Holland, the contract prices for bulbs of the newly introduced and exotic tulip reached extraordinarily high levels before suddenly collapsing to virtually nothing. People weren't buying the tulips because they possessed an intrinsic, functional utility worth the price of a house; they bought them because they were swept up in the collective illusion that the trajectory would go up forever.

[ Real Intrinsic Utility ] <------- Huge Speculative Gap -------> [ Hype / High Market Value ]
  (Fancy Data Processor)                                            (The Omniscient AI Tutor)

The current obsession with running classrooms via generative AI mirrors this historic bubble. We are attributing an almost mystical capability to a tool simply because its output looks dazzling on the surface. Just as the Dutch mistook a fleeting botanical trend for permanent economic foundational wealth, modern educational policymakers are mistaking sophisticated pattern-matching for a replacement for human pedagogy.

Why a Calculator Can't Run a Classroom

Using a calculator to solve an engineering problem is highly effective because math is deterministic. But education is fundamentally relational, adaptive, and non-linear. Relying on an LLM to manage a classroom or dictate a student’s learning pathway is irrational for three distinct reasons:

• The Nuance Blindspot: Human educators continuously read the room. They spot the subtle furrow of a student's brow, sense the emotional deflation behind a wrong answer, and pivot their explanation on the fly based on empathy and shared human experience. An LLM cannot detect or interpret these human variables because it lacks a sensory-motor interface with the real world.
• The Hallucination Hazard: Because calculators of language operate on probability rather than truth, they inevitably hallucinate facts while maintaining an authoritative tone. In a learning environment, exposing developing minds to a system that prioritizes plausibility over truth is pedagogically irresponsible.
• The Erasure of the Learning Process: When classrooms rely too heavily on automated, algorithmic systems, the focus shifts entirely to the final product—getting the "right" answer quickly. True learning, however, is found in the messy friction of drafting, failing, reflecting, and revising.

Keeping the Teacher in the Director's Chair

This is not a call to banish AI from schools entirely. A calculator is an indispensable tool in a math classroom—when placed in the hands of a student who is being guided by a human teacher. It handles the rote, repetitive crunching so that the human mind can focus on high-level conceptual framework and creative problem-solving.

AI can comfortably serve as a digital assistant, helping teachers draft administrative emails or generate raw ideas for worksheets. But it must never be allowed to pilot the ship.

Education is not a data-transmission problem to be optimized by a machine; it is a human relationship. If we hand the core responsibilities of teaching over to statistical calculators, we will eventually watch our speculative educational bubble burst, leaving behind a generation of students managed by algorithms but stripped of genuine critical thought.

To hear Jeff Hawkins break down the biological mechanics of how the brain actually builds knowledge—and why today's AI falls completely short of that framework—take a look at this deep-dive interview on the Thousand Brains Theory of Intelligence. This discussion provides critical scientific context on why today's deep learning architecture operates more like a sophisticated calculating utility than a true thinking mind.