Wiki › Thinking next to AI

AI and Mental Effort

How AI can polarize cognition by either replacing formative effort or amplifying human volition, judgment, and learning.

When machine intelligence becomes abundant, human differentiation may shift from raw cognitive ability toward volition: the desire and willingness to undertake difficult mental work. The Atlantic article “The People Who Will Thrive in the AI Age,” amplified by Peter Steinberger and Farhan Thawar, frames this as a relationship to effort rather than intelligence alone. Psychologists' “need for cognition” describes whether someone intrinsically enjoys demanding thought; it correlates with intelligence but is not identical to it.^{1 2}

Three user trajectories

The article proposes three archetypes, best treated as a conceptual model rather than a validated population taxonomy:

• Productive passengers use AI to reduce thinking. Their immediate output rises, but repeated offloading may weaken memory, knowledge, and unaided performance.
• Reluctant optimizers recognize that danger but gradually prioritize frictionless output under deadline pressure. They risk valuing the product while losing the apprenticeship that forms judgment, curiosity, and independent capability.
• Mental marathoners deliberately preserve difficulty. They use AI to expand agency, challenge their thinking, and attempt more ambitious work rather than to eliminate the effort through which expertise develops.

This model sharpens Cognitive Surrender: the decisive boundary is not whether AI participated, but whether the human formed an independent view, retained judgment, and became more capable through the interaction.

Cognitive polarization

The article predicts a possible cognitive polarization. Some people will use AI to think more and become increasingly capable; others will use it to think less and become increasingly dependent. If this feedback loop compounds, inequality may concern not only income or employment, as in AI Labor Market, but the capacity to learn, judge, resist persuasion, and act without machine mediation.

This is a directional hypothesis, not a settled forecast. The article combines empirical studies, analogies, and moral argument, and some cited measurements have narrower scopes than its broad societal conclusion. Existing evidence also complicates a simple “delegation causes decline” story: perceived ease does not always equal measured productivity, as Speedup Illusion shows, while some user surveys report learning even among heavy delegators. Task design, verification, prior expertise, and whether AI acts as answer machine or tutor are likely important moderators.

Cultivation versus optimization

The central conflict is between two postures:

• Optimization: maximize output while minimizing friction.
• Cultivation: accept useful friction because the process develops the person performing it.

AI systems are usually optimized to remove friction, but learning often requires a zone of difficulty: hard enough to demand effort, not so hard as to overwhelm. Eliminating all friction can produce “competence without apprenticeship”—a plausible result without the understanding and identity formed by earning it.

Practices that preserve agency

• Ask for hints, not finished answers when the purpose is learning.
• Start independently: write an initial analysis or prediction before consulting the model.
• Use AI as a challenger: ask it to find weaknesses, counterarguments, and missing evidence.
• Alternate assisted and unaided work so independent capability remains exercised.
• Separate rote from formative work: delegate low-value mechanics, but retain tasks whose difficulty builds judgment or voice.
• Ask for thinkers rather than thinking: use the model as a librarian that maps existing schools of thought, then engage the underlying sources yourself.
• Design tutors, not oracles: systems should guide users through structured reasoning instead of merely returning conclusions.

These practices complement AI-Assisted Software Development workflows that require the human to understand diffs, reconstruct design choices, and verify runtime evidence rather than merely approve plausible output.

Human advantage beyond intelligence

The article's deepest claim is anthropological: if machines become more intelligent than humans, intelligence alone cannot define human distinctiveness. AI can calculate and synthesize, but it does not possess a lived history, biological needs, an order of loves, or a future self it longs to become. Human advantage therefore lies in aspiration—deciding what is worth wanting—and in the propulsion to endure difficulty while moving toward it.

The practical implication is that schools and organizations should cultivate autonomy, competence, relatedness, admiration, apprenticeship, curiosity, and ambitious missions. In an AI-rich environment, the scarce resource may be less the availability of answers than the formation of people who genuinely want to pursue difficult and worthwhile questions.

Curriculum sequencing

Paul Alivisatos's UChicago proposal turns cultivation into a curriculum sequence: protect foundational writing and reasoning without machines, then deliberately introduce AI and computational tools. His September 2026 column describes forthcoming undergraduate changes and announced law-school sequencing, not evidence of superior learning outcomes.³