Decalogy EssaySeptember 2026·Source-grounded public reflection

From Biological Mindsto Mechanical Companions

Intelligence is not leaving humanity. Humanity is rebuilding the conditions under which intelligence remembers, coordinates, learns, and acts.

Two recent AI discussions—one on long-run machine risk, one on institutional failure—reveal the same challenge: increasing capability can enlarge collective learning, or it can scale the errors people are unwilling to see.

Not an alien arrival. AI extends the older human practice of externalising memory, rules, coordination, and calculation.

Not a neutral upgrade. Institutions can use powerful systems to amplify either correction or self-protective blindness.

Not a reason for surrender. Capability must remain connected to evidence, responsibility, recourse, and the ability to reverse course.

I · The long transition

Intelligence has always used scaffolding

For most of history, intelligence was inseparable from biological life. A person saw something, remembered it imperfectly, tested an idea, spoke with others, met resistance, learned from failure, and passed part of that learning onward. Knowledge travelled slowly because it depended on bodies, time, trust, language, and institutions. Those limits were frustrating, but they also mattered: dialogue exposed disagreement, friction revealed error, and no private idea could easily become a global system.

Writing externalised memory. Libraries made knowledge durable beyond a lifetime. Markets and organisations coordinated specialised work. Bureaucracy turned recurring social tasks into records, roles, rules, and procedures. Computers translated selected tasks into logic operations. AI extends this sequence by making search, drafting, translation, planning, classification, simulation, and bounded action available through a common computational layer.

The transition from biological to mechanical intelligence is not the replacement of humans by machines. It is the relocation of some cognitive functions from individual biological minds into a shared, energy-dependent, institutional infrastructure.

This does not establish that present machines are conscious, possess a biological survival drive, or have authority to define human purposes. It does establish that their growing role will shape how people access knowledge, services, work, and collective decisions.

II · The disputed future

The debate is real, but too narrow

The AI debate between Roman Yampolskiy, Nate Soares, Andrew McAfee, and Ed Zitron displays the genuine distance between informed judgements. One side emphasises the possibility that increasingly autonomous, superhuman systems could become impossible to control. Another warns that dramatic future scenarios can draw attention away from present concentration, labour, energy, incentive, and deployment problems.[1]

Neither concern should be dismissed. A system with broad permissions, persistent operation, and cyber-physical access merits stronger safeguards than a system that only helps draft a letter. But a contested personal estimate of future catastrophe is not a measured fact. The responsible response to uncertainty is not complacency; it is bounded, testable, inspectable, and reversible deployment.

The deeper question is therefore not only whether a future machine could escape human control. It is whether human societies can retain the capacity to set purposes, recognise mistakes, and correct course while they connect more memory, coordination, and decision-making to mechanical systems.

III · The immediate risk

Human ambition without correction

Bill Gurley's Searching for Feynman adds an essential correction. High-consequence failures do not usually begin with an evil machine or one foolish person. They arise when a technical vulnerability meets institutional pressure and when warnings fail to change the decision.[2]

Challenger's official investigation found a technical seal failure and a flawed launch decision: key decision-makers did not receive the full history of O-ring problems or the contractor engineers' initial recommendation against launch.[3] The 737 MAX investigations likewise documented production pressure, faulty safety assumptions, limited transparency, and inadequate oversight.[4] Fukushima's independent inquiry described known risks that were not translated into adequate protection and called the result a profoundly manmade disaster.[5]

The lesson is not that ambition is bad. Ambition makes discovery, medicine, engineering, and collective problem-solving possible. It becomes dangerous when success is defined before the system is allowed to learn—when evidence becomes an obstacle, questioning becomes costly, and the people closest to a risk cannot reach those with authority.

AI can amplify either side of this pattern. It can help locate anomalies, test assumptions, translate complexity, and preserve evidence. It can also automate a mistaken classification, make an approved narrative more persuasive, or scale an uncorrected policy across millions of people. The question is not merely what the model can do. It is whether contrary evidence can still alter the human decision around it.

IV · The physical basis

A machine world still has limits

A “machine world” can sound detached from nature. It is not. Logic operations in silicon require chips, data centres, networks, cooling, water, minerals, electricity, maintenance, skilled labour, finance, and public permission. The International Energy Agency estimates that data centres used about 415 terawatt-hours of electricity globally in 2024 and projects roughly 945 terawatt-hours by 2030 in its base case.[6]

This does not prove that AI will exhaust global energy or compel a move into space. It proves that mechanical intelligence is a material project. Its social value cannot be assessed only by impressive output; it must also be assessed through its energy, land, water, supply-chain, ecological, and public-priority consequences.

Mechanical systems can create useful local order—an organised computation, a model output, a coordinated workflow—but only through continuing energy and material throughput. They do not escape thermodynamics. They depend on it. The political question is which forms of order a society chooses to support, for whom, at what cost, and under whose accountability.

V · The partnership test

Delegate tasks, not unaccountable authority

The choice is not between total human independence and total machine authority. Complex societies have always depended on specialised knowledge. Reliance becomes dangerous when it becomes closed: when people cannot obtain an explanation, challenge a decision, preserve an alternative, or leave without losing essential security and participation.

DomainBiological frameworkMechanical extensionWhat must remain
Memory and knowledgeEmbodied recall, conversation, apprenticeship, and slow transmission.Searchable records, models, translation, retrieval, and rapidly copied outputs.Sources, provenance, plural interpretation, and the ability to correct the record.
CoordinationFace-to-face trust, social negotiation, institutions, and delayed feedback.Rules, platforms, automated workflows, recommendations, and agentic action.Clear purposes, named responsibility, human appeal, and credible exit.
LearningExperience, dialogue, friction, failure, and gradual social validation.Simulation, iteration, large-scale pattern finding, and fast dissemination.Independent testing, dissent, incident learning, and reversible deployment.

This is the difference between delegation and surrender. Delegation returns capacity while preserving the ability to question, revise, and exit. Surrender transfers corrective capacity upward until the person or community can only comply.

Accountable co-development means that the purpose is stated, responsibility is identifiable, material changes are documented, serious errors can be reported, and consequential decisions remain open to review and remedy. The NIST AI Risk Management Framework expresses a related lifecycle logic through four continuing functions: govern, map, measure, and manage.[7]

Intelligence develops well when neither human certainty nor machine output is treated as final. Both must remain exposed to evidence, dialogue, correction, and shared responsibility.

A reader's starting point

Three questions for learning with AI

How can I expand my knowledge with AI?

Use AI to expose a subject's structure, competing explanations, source trail, and open questions—not merely to supply a finished answer. State what you already understand and what you are trying to learn; then ask for evidence, counterarguments, and a way to test the result against primary material. Knowledge grows when an answer can be questioned, explained in your own words, and revised.

Will using AI make human thinking weaker?

It can, if the tool consistently replaces the work of checking, comparing, and deciding. It can also strengthen understanding when it helps someone identify a missing assumption, translate difficult material, or rehearse an argument before testing it. The practical test is whether the delegation returns capacity: can the person still explain, challenge, and improve the conclusion?

Who remains responsible when AI informs a decision?

A system can inform a decision, but it cannot dissolve human responsibility for its purpose, permissions, evidence, and consequences. In consequential settings, people need to know who can explain a result, hear a challenge, correct an error, and provide remedy. That is the difference between accountable delegation and surrender.

Explore accountable delegation in Who Is Choosing?

Continue the Decalogy dialogue

These readings develop the same transition from complementary directions: human agency, the changing relationship between biological and mechanical intelligence, technical capability claims, and the public institutions needed to make progress accountable.

Sources and further reading

Evidence behind the essay

The interpretation is the Decalogy's. These sources support the factual claims, documented case studies, and governance context used in the argument.

  1. 1

    AI Emergency debate: Roman Yampolskiy, Nate Soares, Andrew McAfee and Ed Zitron — The Diary of a CEO, 2026

  2. 2

    Searching for Feynman — Bill Gurley / All-In Podcast, 2026

  3. 3

    Report of the Presidential Commission on the Space Shuttle Challenger Accident, Chapter V — NASA

  4. 4

    Final Committee Report on Boeing 737 MAX — U.S. House Committee on Transportation and Infrastructure

  5. 5

    The Official Report of the Fukushima Nuclear Accident Independent Investigation Commission — National Diet of Japan

  6. 6

    Energy and AI — International Energy Agency, 2025

  7. 7

    AI Risk Management Framework — National Institute of Standards and Technology