While the tech industry obsesses over making AI more like a computer, a new essay argues this fundamental metaphor is broken. Benjamin Riley, writing for The Verge, posits that our brains are not Turing machines but dynamic feedback-control systems evolved over millennia. This distinction matters for developers and infrastructure builders because current AI tools, designed around input-output logic, are actively degrading human cognition rather than augmenting it. Riley compares AI to 'cognitive junk food,' warning that while convenient, it clogs the mental arteries we need for sustainable learning and agency.

The Flawed Computational Model

The prevailing view, championed by figures like Google’s Demis Hassabis and Elon Musk, treats the brain as a biological approximation of a computer. This model suggests thought is a linear algorithm: input, computation, output. However, neuroscientist Paul Cisek at the University of Montreal argues this is a warped mirror. Cisek’s research indicates brains are feedback-control systems that adjust actions to maintain desired states in the environment. A baseball outfielder doesn’t calculate gravity and velocity to catch a fly ball; they use a heuristic to keep the ball in their visual field. AI tools that ignore this dynamic, embodied nature fail to support how humans actually process and retain information.

Education and the Collapse of Critical Thinking

The practical fallout of this mismatch is most visible in education. OpenAI VP Leah Belsky claimed in July 2025 that ChatGPT had become the 'world’s largest learning platform,' with learners comprising half of its 900 million monthly users. Yet, evidence suggests this convenience comes at a cost. A recent study from China found that thousands of students stopped doing homework entirely after adopting AI tools, resulting in substantial harm to learning outcomes. François Chollet describes this as 'cognitive automation,' where software encodes human abstractions, allowing users to bypass the effortful thinking required to build durable knowledge. The result is a habituation to relaxed judgment, weakening the critical discussion and verification practices that institutions like schools rely on.

Policy Shifts and Growing Resistance

In response to these cognitive risks, policy levers are finally being pulled. Norway recently banned nearly all AI use in schools for students under 13, and major US districts like Los Angeles and New York City have implemented similar restrictions. Teachers’ unions are calling for prohibitions on chatbots in elementary schools, recognizing that early reliance on AI undermines the development of autonomous reasoning. On the ground, resistance is growing among students and graduates. Videos of college graduates booing commencement speakers who praise AI signal a cultural pushback against the erosion of 'raw inquiry' and human imperfection. The Oberlin Luddite Club’s open letter to their university president encapsulates this sentiment, prioritizing 'self-actualization' over technological integration.

Key Takeaways

  • The computational model of the mind is insufficient for understanding human cognition, leading to poorly designed AI tools.
  • AI in education is causing 'cognitive delegation,' where students bypass critical thinking and homework, harming long-term retention.
  • Governments and school districts in Norway, Los Angeles, and New York City are implementing bans on AI for younger students to preserve cognitive health.
  • Developers must recognize AI as a cultural technology that risks dismantling the social institutions necessary for human learning.

The Bottom Line

We are building AI infrastructure on a false premise. Until developers stop treating the brain as a static input-output machine and acknowledge it as a dynamic feedback system, AI will remain a cognitive junk food—tasty, but bad for our mental health.