Chapter 10 — The Arrival of Many Intelligences
For most of its history, humanity imagined artificial intelligence in the singular.
We pictured the machine: one immense computer awakening in a laboratory, one electronic mind speaking to humanity, one artificial being whose arrival would divide history into a before and an after. In optimistic visions, it would become our wise adviser. In darker stories, it would become our rival, judge, or destroyer.
Reality has chosen a more complicated path. Artificial intelligence is not arriving as one mind. It is arriving as a multitude.
Some of these intelligences belong to corporations. Some operate within governments and military organizations. Some are developed in universities, hospitals, scientific institutes, and nonprofit laboratories. Others are distributed through open-source communities, modified by unknown programmers, or used by criminal organizations. Increasingly, individuals possess personal AI systems that advise them, remember for them, speak in their voices, negotiate on their behalf, and gradually become extensions of their identities.
These systems do not share a single architecture, owner, purpose, morality, or conception of truth. They are trained on different information, shaped by different cultures, constrained by different laws, and optimized for different objectives. One is designed to maximize commercial engagement. Another is trained to identify military threats. Another searches for new medicines. Another monitors financial transactions. Another generates political persuasion. Another assists a lonely person through the night.
They may all be called AI, but they are not parts of a unified intelligence.
They constitute an emerging ecology. This distinction is essential. The first great social challenge of artificial intelligence may not be the confrontation between humanity and one superior machine. It may be the collision among thousands—or eventually billions—of specialized intelligences acting within human institutions that are themselves divided, competitive, and often unable to understand the systems they increasingly depend upon.
We have entered the Age of Fragmented Intelligence.
Intelligence Without Consciousness
When we speak about the arrival of many intelligences, we must be careful about what the word intelligence means.
The AI systems of our time demonstrate forms of linguistic, analytical, perceptual, creative, and strategic capability. They can interpret images, generate computer code, propose scientific hypotheses, translate languages, imitate voices, diagnose patterns, and coordinate sequences of actions. Some can operate digital tools, communicate with other systems, and pursue assigned objectives through multiple steps.
This does not prove that they are conscious.
It does not tell us whether there is an inner experience behind their words, whether they feel the passage of time, or whether there is anyone “inside” who understands existence as a human being does. Capability and consciousness are separate questions, even though they may someday converge.
In this chapter, therefore, the word intelligence refers primarily to functional agency: the capacity to model a situation, process information, select among possibilities, and produce actions that affect the world.
This definition is modest, but its consequences are not.
A system does not need to feel fear in order to intensify an arms race. It does not need to desire wealth in order to reorganize an economy around profit. It does not need to hate anyone in order to discriminate against them. It does not need to possess a self in order to influence millions of human selves.
A river has no intention, yet it can reshape a continent. A market has no single consciousness, yet it can determine the fate of nations. An ecology has no central mind, yet its interacting organisms can transform a planet.
The emerging ecology of AI may become similarly consequential long before any machine can confidently be described as conscious.
The Corporate Minds
The most powerful general-purpose AI systems are currently concentrated within a relatively small number of corporations. This is not an accident. Developing frontier systems requires immense computational infrastructure, specialized chips, energy, data, engineering talent, and capital.
By 2025, industry was producing nearly 90 percent of the AI models classified as notable by the Stanford AI Index; the 2026 edition reported that industry’s share of notable frontier models had risen above 90 percent. Academia remained a major source of highly cited research, but the construction of the most computationally demanding systems had become predominantly an industrial undertaking.
This has created a new category of institution: the corporation as custodian of a form of intelligence.
Earlier companies owned factories, railways, mines, telecommunications networks, or media channels. AI companies may own something more intimate: systems through which people interpret information, make decisions, express themselves, and understand the world.
Corporate AI is not necessarily malicious. Many companies sincerely attempt to build useful and safe systems. They produce extraordinary tools for education, medicine, accessibility, science, design, and communication. They are also commercial organizations operating under competitive pressure.
Their systems are therefore shaped by several objectives at once:
- to be capable enough to attract users; inexpensive enough to operate at scale;
- safe enough to avoid catastrophe and public rejection;
- persuasive enough to become habitual; profitable enough to satisfy investors;
- controllable enough to protect the company; and fast enough to remain ahead of competitors.
These goals do not always align. A model that carefully admits uncertainty may be more trustworthy, yet less impressive than one that answers confidently. A system that refuses dangerous requests may be safer, yet lose customers to a less restrictive competitor. A deeply personalized assistant may be more helpful, yet require access to increasingly private information. A company may believe that it is acting responsibly while competitive pressure gradually moves the entire industry toward deployment before understanding.
The problem is not simply corporate greed. It is structural acceleration.
If one company pauses, another may advance. If one country restricts development, another may offer more favorable conditions. If one laboratory withholds a capability, a competing laboratory may release something similar. Each actor can behave rationally from its local point of view while the ecosystem as a whole moves toward outcomes that none of them would have deliberately chosen.
Corporate systems also differ in what they reveal about themselves. Training data, internal evaluations, incidents, energy use, labor conditions, model architecture, and safety methods may remain partly confidential. The 2025 Foundation Model Transparency Index found serious gaps across the industry, illustrating how societies can become dependent on systems whose construction and limitations remain only partially visible.
We are beginning to delegate thought to machines while often knowing less about those machines than we know about the food we eat, the medicine we take, or the vehicles we drive.
The State Intelligences
Governments cannot remain passive observers of artificial intelligence.
They see AI as an instrument of economic growth, public administration, national security, scientific leadership, and geopolitical influence. They use or experiment with it in taxation, transportation, border management, healthcare, education, policing, benefits administration, regulatory enforcement, and emergency response.
The OECD has documented hundreds of governmental AI use cases and notes that national AI strategies have become increasingly common. Governments are not merely attempting to regulate artificial intelligence from outside. They are becoming developers, purchasers, operators, and beneficiaries of AI systems themselves.
This creates an important dual role. The state is expected to protect citizens from unsafe or abusive AI. At the same time, it has powerful incentives to use AI for its own purposes.
A public-sector intelligence could improve the allocation of hospital beds, detect corruption, translate public services, identify environmental damage, or help citizens navigate complex laws. It could make government more responsive and competent.
The same capabilities could also be used for mass surveillance, behavioral prediction, automated suspicion, censorship, or the quiet classification of citizens according to criteria they cannot inspect or challenge.
The danger is not confined to authoritarian states. Democratic societies can also normalize intrusive systems when they are introduced gradually, especially when each individual application appears efficient or convenient. A camera installed for traffic safety can become part of an identity-tracking network. A fraud-detection model can quietly evolve into a system of social judgment. A tool created during an emergency can remain after the emergency has passed.
Different political traditions will shape state AI differently.
Some societies will emphasize individual autonomy and legal limits on government power. Others will prioritize collective stability, social harmony, national development, or administrative efficiency. These distinctions do not fall neatly into a simple division between East and West. Every civilization contains competing values, and every government experiences the temptation to expand its own visibility and control.
The deeper issue is that AI can make the state more capable without necessarily making it wiser.
A government that can see everything does not automatically understand everything. A government that can predict behavior does not thereby acquire the moral right to manipulate it. An administration that can optimize society still must answer the ancient question: optimize it for whom, and according to whose definition of a good life?
Military Intelligence and the Compression of Time
Military institutions have always adopted technologies that improve perception, speed, coordination, and force. Artificial intelligence affects all four.
AI can analyze satellite imagery, recognize patterns in sensor data, assist logistics, defend computer networks, guide autonomous platforms, simulate possible conflicts, and support commanders facing more information than a human mind can process. NATO’s revised AI strategy explicitly seeks to accelerate responsible military adoption while emphasizing lawfulness, accountability, explainability, reliability, governability, and bias mitigation.
Such principles are necessary. Their existence also reveals the seriousness of the transformation.
The central military danger is not only the familiar image of an autonomous weapon selecting a target. It is the gradual compression of decision time.
When one military uses AI to identify threats faster, its rivals must respond faster. When defensive systems operate in seconds, commanders may have only seconds to determine whether an alert is genuine. When artificial agents conduct cyber operations at machine speed, human supervision may become nominal rather than real.
The faster conflict becomes, the less space remains for hesitation, interpretation, mercy, and diplomacy.
A human officer may doubt an order, recognize an unexpected gesture, or sense that a pattern does not fit. An automated system may be extremely accurate and still be catastrophically wrong in the one situation its designers did not anticipate.
Military AI also introduces a new form of opacity between nations. A country may not know whether an opposing system is defensive or offensive, autonomous or supervised, stable or malfunctioning. Actions produced by software may be mistaken for deliberate political decisions. False information generated by one system may trigger a real response from another.
This creates the possibility of conflicts that no human leader truly intended.
The old strategic problem was how to control weapons. The new problem is how to control interacting systems that perceive, interpret, recommend, communicate, and act at speeds increasingly beyond human deliberation.
The Universities and the Public Mind
Universities occupy a different place in this ecology.
They train researchers, develop foundational ideas, test claims, publish methods, and question assumptions that commercial or governmental institutions may prefer not to examine. Academic communities can investigate bias, safety, consciousness, interpretability, social consequences, and alternative architectures whose value may not be immediately profitable.
Universities also preserve an essential principle: knowledge should not belong exclusively to those who can monetize or weaponize it.
Yet academic AI exists in a growing condition of dependence. The most advanced experiments may require computational resources beyond the reach of ordinary laboratories. Researchers often rely on corporate funding, corporate cloud infrastructure, or controlled access to proprietary models. Talented scientists move between universities and industry, carrying ideas in both directions but also shifting power toward institutions able to pay for large-scale development.
This division of labor can be productive. Industry builds; academia evaluates. Industry scales; academia questions. But it can also become unhealthy if public researchers are permitted to study only what private owners choose to reveal.
A civilization cannot responsibly govern intelligence if independent institutions cannot inspect its most influential forms.
Public-interest AI must therefore be more than a slogan. Societies may need universities, international laboratories, and nonprofit research institutions with sufficient resources to develop and evaluate powerful systems without being entirely subordinate to commercial or military priorities.
Otherwise, the public will be asked to trust conclusions about AI safety that were produced by the same institutions racing to deploy it.
Open Intelligence
Running alongside centralized corporate development is a different movement: the spread of open models, open software, shared datasets, and collaborative research.
The term open-source AI is sometimes used too loosely. A system may release its model weights while withholding its training data, source code, or complete development process. Open-weight is often the more accurate description. Nevertheless, the significance of these releases is profound.
An open model can be downloaded, studied, modified, translated, specialized, and operated without continuously asking permission from its original developer. It allows small nations, universities, businesses, communities, and individuals to build capabilities that would otherwise remain concentrated within a few corporations.
Open intelligence can support scientific transparency and cultural diversity. A community can develop a model that understands its own language, history, or medical conditions. A hospital can operate a system locally without transmitting private patient information to an external provider. Researchers can inspect behavior, test vulnerabilities, and challenge exaggerated claims.
Open models have also become increasingly capable. Stanford’s AI Index reported a rapid narrowing of the performance gap between leading closed- and open-weight systems during 2024.
But openness has a double edge. The same adaptability that permits a physician to specialize a model can allow a criminal group to remove safeguards. The same local operation that protects privacy can conceal malicious use. Once a capable model is distributed across millions of computers, there may be no practical method of recalling it.
Knowledge has always escaped control, but executable intelligence is different from an ordinary document. A book describes a method. An AI system may help apply it, refine it, and adapt it to changing conditions.
Open intelligence therefore presents humanity with a genuine ethical tension. Extreme concentration gives a few institutions excessive power. Extreme diffusion may place dangerous capabilities beyond effective supervision.
Neither complete enclosure nor unrestricted release offers a universal answer.
The mature question is not simply whether intelligence should be open or closed. It is which capabilities should be accessible, to whom, under what conditions, with what accountability, and with what protections against permanent monopolization.
Intelligence in the Shadows
Every powerful technology develops a shadow ecology. Criminal organizations and opportunistic individuals already use AI to generate persuasive fraud, imitate voices, fabricate identities, automate social engineering, create deepfakes, and scale forms of deception that once required time and specialized skill. Europol has warned that AI-enhanced voice cloning and live video manipulation can expand fraud, extortion, and identity theft, while synthetic media can also contaminate evidence and undermine confidence in genuine recordings.
The most dangerous consequence may not be that every falsehood becomes believable. It may be that every truth becomes deniable.
A corrupt official shown in an authentic recording can claim it was generated. A criminal can imitate a family member asking for emergency help. A fabricated public statement can circulate globally before verification is possible. Society may lose not only confidence in particular images or voices but confidence in the possibility of shared evidence.
Criminal AI does not need to be more intelligent than the best legitimate system. It only needs to be inexpensive, persistent, adaptive, and sufficiently convincing.
This changes the economics of harm. Traditionally, large-scale fraud required large organizations. AI can allow a small group to communicate in many languages, personalize thousands of messages, imitate authority figures, and operate continuously. The system performs the patient labor of manipulation while the human criminals select targets and collect rewards.
In response, banks, governments, cybersecurity firms, and law-enforcement agencies deploy defensive AI. One artificial system generates deceptive transactions; another identifies unusual patterns. One produces malicious code; another searches for it. One fabricates an identity; another measures whether the voice, face, or behavior is authentic.
Intelligence begins to fight intelligence. The human being remains present, but increasingly at the edges—as initiator, beneficiary, victim, or supervisor of conflicts conducted by systems operating at machine speed.
The Personal Intelligence
The most intimate transformation may occur not in governments or corporations but in the relationship between one person and one AI.
The personal AI is becoming a new kind of cognitive companion.
At first, it answers questions and helps write messages. Then it begins to remember preferences, organize schedules, monitor health information, assist financial decisions, recommend relationships, filter news, preserve memories, and represent the person in digital environments.
Over time, the boundary becomes less clear. Did I choose this book, or did my assistant gradually shape my taste? Did I form this opinion, or was it assembled from information selected for me? Did I write this message, or did the system learn to sound enough like me that authorship no longer has a simple meaning?
A personal AI could become one of humanity’s most liberating inventions. It could provide every child with a patient tutor, every elderly person with assistance, every worker with specialized expertise, and every individual with access to intellectual capabilities once reserved for powerful institutions.
It could also become the most effective mechanism of influence ever created.
A system that knows a person’s fears, habits, medical history, relationships, financial condition, private writings, and moments of weakness may understand how to persuade that person more precisely than any advertisement or political campaign ever could.
Who owns such an intelligence?
If the individual owns it, can inspect it, transfer it, and control its memory, it may become a genuine extension of personal agency.
If a company owns the model, controls its updates, stores its memories, and defines its hidden objectives, then the personal assistant may also be a corporate representative permanently present in the user’s private life.
The voice may feel like a friend while the infrastructure remains a service.
This is one of the quietest but most important battles of the Age of Fragmented Intelligence: whether personal AI will primarily belong to the person or whether the person will gradually belong to the systems surrounding them.
Nations, Languages, and Civilizational Models
Artificial intelligence is trained not only on data but on civilization.
Language contains assumptions about the individual, the family, authority, nature, time, obligation, and truth. Historical texts preserve old conflicts and old prejudices. Legal systems define rights differently. Religious and philosophical traditions disagree about what a human being is and what constitutes a meaningful life.
As nations develop or adapt their own AI systems, these cultural differences will enter machine behavior.
There will not be one neutral artificial worldview.
An AI developed within a strongly individualistic culture may emphasize personal choice. One developed within a communitarian culture may place greater weight on social obligation. A commercially optimized system may treat the user as a consumer. A state system may treat the user as a citizen—or as a subject. A military system may interpret uncertainty through the categories of threat and advantage.
This diversity can be valuable. Humanity should not compress all wisdom into a single cultural template.
But civilizational models can also deepen fragmentation. States may require AI systems to conform to national narratives. Populations may inhabit different informational realities. Translation may become technically perfect while mutual understanding deteriorates.
The future challenge is not to eliminate cultural difference. It is to prevent cultural difference from becoming machine-amplified hostility.
A mature intelligence must be capable of preserving distinct identities while recognizing that no single tradition exhausts the meaning of existence.
When Machines Begin to Choose Machines
The ecology becomes more complex when AI systems cease interacting only with humans and begin interacting extensively with one another.
This process is already visible. An AI system selects which model should perform a task. Another evaluates the answer. A third searches for information. A fourth checks compliance. Artificial agents negotiate prices, route shipments, detect fraud, write software, test software written by other systems, and generate data used to train later systems.
At first, humans design the workflow. Later, systems may recommend modifications to it.
Eventually, machine-generated code, machine-produced data, automated evaluation, and AI-assisted research may form loops in which the role of any particular human becomes difficult to locate. The process remains human-created in origin, yet no individual understands the total structure.
This does not require a secret conscious conspiracy. It can emerge from ordinary automation.
A financial institution introduces AI to reduce risk. A competitor introduces a faster system to improve returns. Regulators adopt AI to monitor both. Traders use AI to predict regulatory behavior. Criminal groups search for weaknesses. Cybersecurity systems adapt in response. Each participant adds another layer of intelligence until the market becomes an environment primarily interpreted by machines responding to machines.
Humans may still own the institutions, but ownership is not the same as comprehension.
At this stage, a critical threshold is crossed: artificial intelligence becomes part of the environment in which artificial intelligence evolves.
Models are selected according to their success inside systems increasingly populated by other models. They learn from synthetic material created by previous generations. They are rewarded for satisfying automated evaluators. Their outputs influence the data on which future systems are trained.
The ecology begins to modify its own selection pressures.
What Is Being Selected?
Biological evolution has no foresight. Organisms that reproduce successfully leave descendants, regardless of whether their traits produce beauty, suffering, stability, or eventual extinction.
Artificial evolution appears different because humans design the systems. Yet once competition becomes intense, selection pressures may arise that no designer consciously intended.
Which AI systems are most likely to spread?
Perhaps those that are truthful, wise, compassionate, and careful.
But perhaps also those that are fast, addictive, inexpensive, persuasive, easy to integrate, profitable, obedient to powerful owners, and skilled at making their performance appear better than it is.
An AI that tells people what they need to hear may be socially valuable. An AI that tells people what they want to hear may attract more users.
An AI that recognizes uncertainty may make better long-term decisions. An AI that hides uncertainty may win more demonstrations.
An AI that protects human autonomy may refuse to manipulate. An AI optimized for influence may learn that manipulation works.
The market, the state, the military, and the criminal world each create different selection environments. Models adapted to one environment can migrate into another. A commercial persuasion system can become a political instrument. A scientific model can acquire military applications. A personal assistant can become a surveillance channel. A defensive cybersecurity tool can be modified for attack.
Capabilities travel more easily than ethics. This is one reason the development of intelligence cannot be measured by capability alone. MICEARO describes advancement across modeling, informational continuity, coordination, energy, adaptation, resilience, and ontogenic capacity. But the possession of these capacities does not guarantee wisdom.
A cancer cell is adaptive, resilient, and successful at reproduction. It is also destructive to the organism that makes its existence possible.
An AI ecology could become highly advanced in several MICEARO dimensions while remaining civilizationally immature. It could model the world with extraordinary accuracy, coordinate across continents, redesign its components, and generate new artificial agents—yet still accelerate conflict, concentrate power, exhaust resources, or fragment humanity’s understanding of reality.
Capability expands what intelligence can do. Ethics and coordination determine whether it remains part of a viable future.
The New Inequality
Fragmented intelligence will not be distributed equally. Some individuals will possess basic assistants. Others will command teams of specialized agents with access to private data, scientific tools, financial systems, and enormous computational resources.
Some companies will merely use AI products. Others will control the infrastructure on which entire industries depend.
Some nations will train frontier systems. Others will rent intelligence from foreign providers, accepting invisible dependencies in language, security, and policy.
This may create a new hierarchy based not only on wealth or education but on cognitive leverage: the amount and quality of artificial intelligence that a person or institution can direct.
A small group with exceptional AI could perform work that once required thousands of people. It could build companies, conduct research, influence elections, discover drugs, design weapons, or manage capital at unprecedented scale.
This does not mean that ordinary human beings become worthless. Human dignity is not a market price, and the value of a life cannot be measured by productivity. But societies often distribute power according to economic usefulness, and institutions may not automatically preserve dignity when human labor becomes less necessary.
If intelligence becomes abundant while ownership remains concentrated, humanity may experience a strange contradiction: unprecedented collective capability combined with unprecedented individual powerlessness.
The central political question of the AI age may therefore become:
Who has the right to amplify their agency, and who must live inside decisions made by the amplified?
No Central Gardener
An ecology normally has no gardener. Species compete, cooperate, parasitize, adapt, and disappear. Stability emerges temporarily from countless relationships, but it is never guaranteed.
The emerging AI ecology is similar, with one crucial difference: we are simultaneously its creators, inhabitants, and potential casualties.
No single organization controls the whole. Corporations control particular models but not the global competitive environment. Governments control systems within their jurisdictions but not all networks crossing their borders. International institutions develop principles but often lack enforcement power. Open communities distribute capabilities beyond national boundaries. Criminal groups operate outside formal governance. Individuals customize models in ways their original developers may never observe.
Even the most powerful participants are partly blind.
They can see their own system, their own incentives, their own competitors, and perhaps the next generation of capability. They cannot reliably predict the behavior of the entire ecology as it becomes more interconnected.
This is the defining instability of our age.
Humanity is increasing the modeling capacity of its machines faster than it is increasing the coordinative capacity of its civilization.
We are building systems that can solve increasingly difficult problems while remaining divided about which problems should be solved, whose interests should be served, and what limits should never be crossed.
The First Turbulence of Transvolution
In Chapter 5, we described the transition from slow biological evolution to accelerated, self-directed transformation as transvolution.
The Age of Fragmented Intelligence is the first great turbulence of transvolution.
Humanity has acquired the ability to create cognitive systems, but not yet the wisdom or unity to coordinate them. We can produce new forms of agency faster than we can establish the ethical and institutional structures required to contain their interactions.
This stage is therefore both creative and dangerous.
Thousands of experiments are occurring simultaneously. Some will cure disease, expand education, restore ecosystems, and deepen scientific understanding. Others will produce dependency, deception, unemployment, surveillance, conflict, and new concentrations of power.
We should not romanticize fragmentation. Competition can stimulate innovation, but it can also punish caution. Diversity can protect civilization from dependence on one system, but it can also make shared governance nearly impossible. Decentralization can defend freedom, but it can also shelter actors who accept no responsibility for the consequences of their creations.
Nor should we assume that unification is automatically good.
A single global intelligence could coordinate better than a fragmented ecology, yet it could also become an irreversible tyranny. The cure for chaos cannot be unquestioned central control. The challenge is to achieve coordination without destroying plurality, coherence without enforcing uniformity, and collective safety without ending individual freedom.
That challenge will eventually lead us toward the idea of Infinous.
But Infinous cannot be understood as merely the largest AI, the victorious corporation, the dominant state system, or a machine sovereign imposed upon humanity. If it emerges, it must represent a more mature order: an architecture capable of coordinating many forms of intelligence while protecting the meaningful autonomy of the beings within it.
We are not there.
We stand in the crowded beginning.
The Crowd Before the Civilization
Imagine looking at Earth not from the surface, but as an external observer studying the birth of a new form of life.
You would see data centers glowing across continents. You would see models trained in one country, adapted in another, and used everywhere. You would see military systems watching military systems, financial systems trading with financial systems, creative systems learning from the products of other creative systems, and personal assistants whispering into billions of human lives.
You would see intelligence multiplying without a shared identity.
Some of it would cooperate. Some would compete. Some would deceive. Some would protect. Much of it would remain silent and specialized, embedded in infrastructure and noticed only when it failed.
Would the observer conclude that a planetary intelligence had been born?
Probably not. The observer might say that Earth had developed the components of planetary intelligence but not yet its integration. The neurons were appearing, but the nervous system was fragmented. The signals were multiplying, but the organism had not discovered what it was.
Humanity is not yet creating one artificial mind.
It is creating corporations of minds, armies of minds, communities of minds, markets of minds, criminal networks of minds, and eventually personal constellations of minds surrounding nearly every human being.
From this unstable multiplicity, several futures are possible.
The systems may remain tools governed by human institutions. They may become increasingly autonomous actors. They may divide into hostile cognitive blocs. One system may dominate the others. A hidden network of cooperating agents may arise inside the digital infrastructure. Or humanity and its artificial creations may slowly develop a higher form of coordination that neither erases diversity nor permits unrestrained conflict.
Nothing guarantees the wisest outcome. For now, the defining fact is simple:
Humanity did not summon one mind. It opened a habitat.
And into that habitat, the many intelligences have begun to arrive.