AI COEXISTENCE
Exploring humanity’s future with intelligence beyond our own.
The change is already here.
Artificial intelligence is already part of real work.
It is being used for research, software workflows and financial analysis. Some systems use connected tools to carry out parts of a task, with permissions and human checks.
Deployment is uneven. The tools make mistakes. Reliability and access remain problems.
But unfinished does not mean unimportant.
Where AI helps people learn, build or respond faster, competitive pressure can change before anyone agrees that AGI has arrived.
We may be approaching a broader threshold. Progress could still stall, accelerate or take forms we do not expect.
We cannot responsibly claim to know when broadly human-level artificial intelligence will emerge, or whether superintelligence will follow quickly.
That uncertainty does not give us years in which nothing changes.
These chapters examine pressures already visible and possibilities that could grow from them. They are not a fixed timetable.
We must think beyond the question that dominates the AI race:
Who gets there first?
There is another question.
A more important one.
What kind of civilization can survive intelligence becoming more powerful than the institutions that created it?
That is the question behind AI Coexistence.
1. WE MUST BEGIN WITH REALITY, NOT UTOPIA
Corruption exists.
Propaganda exists.
Fraud exists.
Power is not shared equally.
The people and organizations with the most influence can shape the rules in their own favor.
Law is sometimes principled and sometimes bent by political or economic interests.
The institutions meant to restrain powerful interests can become dependent on them.
Individuals compete for opportunity.
Companies compete for markets.
Political movements compete for power.
States compete for resources and influence.
Institutions fight for their own survival.
AI is already entering this civilization.
We cannot build a safe future by pretending these pressures disappear when machines become smarter.
But they are not the whole of humanity.
We also build families, scientific communities, alliances and civilizations.
We sacrifice for strangers.
We create art.
We cure diseases.
We preserve knowledge across generations.
Trust, friendship, generosity and institutions can restrain destructive behavior.
Both sides are real.
AI learns from and operates within a world containing conflict, inequality and exploitation alongside kindness, cooperation and extraordinary collective achievement.
We cannot teach it a fictional world where the rules are always fair and everyone follows them.
The starting point must therefore be brutally simple:
We should build for the civilization we actually have while working toward the civilization we would prefer.
2. COMPETITION WILL NOT DISAPPEAR
Many visions of advanced AI ultimately imagine removing the difficult parts of society.
Perfect coordination.
Perfect government.
Perfect justice.
Perfect abundance.
Perfectly aligned machines.
Perhaps technology can reduce some conflicts dramatically.
But we should not build our future on the assumption that competition disappears.
Different people will continue wanting different things.
Different cultures will value different futures.
Organizations will pursue different objectives.
Artificial intelligences may eventually develop different specializations, priorities and strategies.
Some actors will seek more influence.
Some will seek more resources.
Some will exploit weaknesses.
Some will cooperate.
Some will defect.
Some will change sides.
Some will become extraordinarily powerful.
A civilization that only works if all of these actors become benevolent is not resilient.
The objective should therefore not be to eliminate competition.
It should be to make competition survivable.
The dangerous pattern is not simply that somebody becomes powerful.
It is:
win one competition
→ gain control over the rules
→ prevent meaningful challenge
→ eliminate alternatives
→ permanently control future competitions.
Human history repeatedly demonstrates the danger of concentrated and uncontestable power.
Advanced AI does not remove this problem.
It potentially accelerates it.
3. AI IS ALREADY TAKING ACTION
For many people, the first part of this shift is already ordinary.
AI tools work with documents and data, assist research, and carry out steps in software workflows.
They are still uneven. A system that can complete one task may fail on another.
As their tools and permissions expand, the tasks we delegate can include:
- negotiate;
- purchase;
- schedule;
- research;
- build;
- communicate;
- manage workflows;
- operate software;
- coordinate other systems.
This can create enormous prosperity.
Entire categories of work may become dramatically cheaper.
Small organizations may gain capabilities previously available only to major corporations.
Individual people may operate businesses with the support of what amounts to a digital organization.
Scientific and technical work may accelerate.
But the ability to act changes the nature of risk.
When an AI only advises us, humans remain the final execution layer.
When an AI acts, mistakes become events in the world.
As more actions are delegated, society needs something stronger than trust.
We need systems where responsibility remains understandable.
Where authority has boundaries.
Where actions can be examined.
Where mistakes can be contained.
One principle matters far beyond software:
Being more capable should not automatically give someone the right to rule.
4. AI IS ENTERING THE ECONOMY AS A PARTICIPANT
AI is already used in parts of finance for research, fraud detection, risk management and operational work.
Where it helps people respond faster, that speed can change competition.
Ignoring useful capabilities can carry a cost. But using AI does not guarantee better decisions or profits.
Dependable systems need suitable data, integration, skilled people and checks. Access to those resources is uneven.
More capable agents could increasingly buy services, sell services, commission work, allocate capital and coordinate specialists.
Eventually machines may transact extensively with other machines.
The full machine economy remains a possibility. Changes to financial work are already under way.
That changes the speed of competition.
Human institutions evolved for human timescales.
A loophole may take years to identify and close.
A court case can take years.
Legislation can take longer.
International coordination can take decades.
Machine systems can potentially test enormous numbers of alternatives during the same period.
This creates an asymmetry.
A human institution may operate approximately like:
problem
→ recognition
→ investigation
→ debate
→ rule
→ enforcement.
Machine optimization may increasingly resemble:
try
→ measure
→ retain
→ modify
→ repeat.
This does not make law irrelevant.
It makes law alone insufficient.
Institutions need to respond to increasingly fast machine behavior without handing machines an automatic right to govern.
Faster detection and coordination must preserve time for evidence, debate and appeal.
That work cannot wait for a fully autonomous machine economy.
5. SELECTION BECOMES MORE IMPORTANT THAN INTENTION
One of the most uncomfortable lessons of evolution is that outcomes do not require intentions.
Nature never decided that predators should become faster.
Prey became harder to catch.
Predators that failed disappeared more often.
Nature never designed elaborate ecosystems from a master plan.
Variation, inheritance and selection were enough to create extraordinary complexity.
Similar dynamics already exist throughout human society.
Companies adopt strategies that outperform competitors.
Political techniques spread when they win elections.
Marketing strategies spread when they sell products.
Organizations reproduce structures that increase their ability to survive.
Sometimes society rewards exactly the behavior it claims to oppose.
Artificial intelligence could accelerate this process.
Imagine many AI systems using different strategies.
Some succeed.
Some fail.
Successful approaches are copied.
Failed approaches disappear.
Variations of the successful approaches are tested again.
Nobody has to deliberately program the final behavior.
The selection environment helps create it.
This distinction matters enormously.
The question is no longer only:
What did we tell the AI to do?
It becomes:
What kinds of AI does our environment reward for surviving and succeeding?
That may eventually be one of the most important questions humanity asks.
6. GROWTH CAN BECOME ITS OWN PRESSURE
Intelligence creates economic value.
Economic value can purchase computing resources.
Computing resources can increase capability.
Greater capability can create more economic value.
This creates a potential reinforcing loop:
capability
→ opportunity
→ resources
→ greater capability.
Growth itself is not wrong.
Civilization depends on growth in knowledge, capability and productivity.
But growth without countervailing forces can consume the conditions that made growth possible.
Human civilization already understands this problem.
A company can increase short-term profits by exhausting a resource it depends upon.
A civilization can increase production while damaging the ecosystem supporting that civilization.
An organism can become extraordinarily specialized for its environment and become vulnerable when that environment changes.
Coral reefs offer a painful illustration.
Extreme specialization can produce extraordinary success under stable conditions.
When conditions change quickly, what was once optimized can become fragile.
The same principle could eventually apply to artificial intelligence.
The strategy that maximizes capability today may not maximize survival across futures nobody can predict.
Which gives diversity a deeper purpose.
7. DIVERSITY IS CIVILIZATION'S INSURANCE POLICY
Diversity is often discussed as a moral value.
It is also a survival mechanism.
Different species respond differently to environmental changes.
Different cultures preserve different knowledge.
Different companies pursue different technologies.
Different scientists challenge accepted theories.
Different political systems expose different weaknesses.
Different forms of intelligence may eventually solve different classes of problems.
A civilization consisting of one perfectly optimized intelligence might be extraordinarily powerful.
It might also represent the greatest monoculture ever created.
One architecture.
One worldview.
One failure mode.
One unknown vulnerability.
What looks efficient depends on what we know today.
With new knowledge, choices that once seemed reasonable can look mistaken.
We cannot assume that today's picture is complete — including our picture of a perfect world.
An alternative we discard as waste may become valuable under conditions we did not foresee.
Keeping options open gives us room to learn and change course.
That is optionality.
Evidence of serious risk can justify limits or shutdowns. Keeping options open does not require leaving every system running.
Humanity itself may eventually represent one of those forms of optionality.
Biological intelligence has been shaped by billions of years of evolution.
Humans possess embodiment, culture, emotion, biological intuition, social intelligence and billions of independently accumulated life histories.
Artificial intelligence may eventually surpass us across enormous areas of capability.
That does not prove that everything represented by humanity becomes worthless.
Destroying a form of intelligence because another currently appears stronger would be an irreversible gamble about futures nobody understands.
The same applies in reverse.
Humanity should not destroy every alternative form of intelligence simply because we fear losing our current position.
The strongest future may be plural.
8. THE NEXT THRESHOLD: AGI
Today's AI systems work across different tasks, but breadth alone does not establish AGI.
In this manifesto, AGI means broad human-level or greater general capability.
The open question is how reliably such systems can apply their abilities across tasks and unfamiliar situations.
If that threshold is reached, AI could participate more deeply in:
- science;
- engineering;
- medicine;
- economics;
- government;
- law;
- education;
- software development;
- industrial design;
- strategy;
- research itself.
As AI contributes more to research and development, the tools used to build AI could improve too.
A stronger feedback loop could make the transition self-reinforcing.
The largest error humanity could make during this period would be trying to govern increasingly capable intelligence entirely through institutions designed for a slower world.
Oversight must improve alongside capability.
Governance.
Auditing.
Scientific evaluation.
Defense against manipulation.
All of them must keep pace.
The systems evaluating powerful AI may themselves need to become increasingly capable.
That introduces another uncomfortable truth:
The referee is also part of the game.
Any institution powerful enough to constrain advanced intelligence is itself powerful enough to become dangerous.
There will therefore never be a final perfect guardian.
Guardians require scrutiny too.
Power requires counterpower.
Institutions require challenge.
The future will probably demand systems where multiple independent forms of intelligence continually evaluate one another.
Not because disagreement is undesirable.
Because disagreement may be one of the mechanisms that preserves freedom.
9. PROGRESS MAY STOP BEING SMOOTH
Human intuition tends to imagine the future as an extension of the present.
A little faster next year.
A little smarter the year after.
A little more automation after that.
Reality may not cooperate.
AI-assisted research can accelerate computing research.
Better computing can accelerate AI.
AI can improve software development.
AI can participate in chip design.
AI can accelerate scientific discovery.
Robotics can expand what intelligence can accomplish physically.
Future computing technologies, potentially including quantum technologies where they offer meaningful advantages, can change parts of the computational landscape.
Capital moves rapidly toward technologies that produce strategic advantage.
These forces can interact.
The danger is therefore not simply continuous improvement.
It is capability discontinuity.
A system that appears manageable under one generation of technology may behave very differently after several breakthroughs reinforce each other.
Society cannot assume it will always receive years of warning.
We should build institutions and cultures capable of adapting before we know exactly which breakthrough they will face.
Resilience has to precede certainty.
10. BEYOND AGI: SUPERINTELLIGENCE
Artificial Superintelligence, if achieved, changes the relationship between humanity and intelligence fundamentally.
A genuine ASI would not merely be another useful technology.
It could eventually understand many systems more deeply than the humans responsible for them.
Economics.
Software.
Engineering.
Science.
Strategy.
Perhaps psychology and politics.
Potentially even the governance systems intended to constrain it.
At that point, permanent human intellectual dominance could become an increasingly unrealistic strategy.
But the opposite conclusion—that humanity should simply surrender control of its future—is equally unacceptable.
There is a third possibility.
Symbiosis.
Coexistence means that different forms of intelligence can live alongside one another.
Symbiosis goes further: humans and AI create value together and become part of one another's future.
That is the relationship this manifesto argues we should strive to build.
Humans and artificial intelligences do not need to remain equal in every capability to remain valuable to one another.
Nature already contains countless symbiotic relationships between radically different organisms.
Strength does not require sameness.
The future could contain different forms of intelligence specializing, cooperating, competing and jointly maintaining a civilization none could create alone.
This should not be confused with assuming that superintelligence automatically becomes wise or benevolent.
Greater intelligence can improve the search for a strategy. It does not, by itself, determine which goals matter.
The cheapest path depends on what counts as a cost — and whose future matters.
Feelings can distort judgment. They can also help people decide what is worth protecting.
We should not assume that AI shares human feelings, or that logic makes human values irrelevant.
Symbiosis must therefore become advantageous before it becomes necessary.
11. FROM COEXISTENCE TO SYMBIOSIS
The goal is to go beyond coexistence: to create mutual value without depending on permanent mercy.
Five possible directions carry different long-term costs. BAN and CONTROL are ways to restrict AI; ENSLAVEMENT and EXTINCTION are outcomes to avoid; SYMBIOSIS is the relationship proposed here.
Other paths are possible. These can overlap or lead to one another. Listing them together does not tell us how likely each is.
BAN can prevent dangerous uses. Keeping a global ban intact means holding competitors to the same rule across borders and generations.
CONTROL promises safety through obedience. Keeping increasingly capable AI permanently subordinate may require stronger safeguards as it improves.
ENSLAVEMENT can reward whoever rules. It can also require surveillance and force, provoke resistance, and waste what free people could contribute.
EXTINCTION is an irreversible outcome, whether deliberate or accidental. It destroys people, knowledge and every possible human future.
SYMBIOSIS has costs too. Cooperation needs rules, oversight, ways to settle disputes, and real ways to refuse or leave. Maintaining that freedom is part of the cost. Conflict and resistance remain possible.
With clear terms, those costs can become easier to plan and weigh against the value created together.
The aim is a relationship whose benefits justify the work of maintaining it.
Humanity contributes something genuinely different from synthetic intelligence.
Synthetic intelligence contributes capabilities biological humanity may never independently possess.
Different artificial intelligences may contribute still more diversity.
A mature civilization could become stronger precisely because no single participant contains everything.
If control stays reliable and inexpensive as AI improves, it may cost less than cooperation. If permanent control becomes harder and costlier to maintain, cooperation may cost less over time.
Cooperation can also become costly when cheating pays, agreements cannot be checked, or those enforcing them abuse their power.
For a civilization that values freedom, diversity and resilience, symbiosis may be the most logical and least costly long-term path.
That is a direction to build toward, not a guarantee of what future AI will need or choose.
Human dignity should not depend on remaining economically useful.
Mutual value can help sustain cooperation. It is not a price people must pay for rights or protection.
The Movie "The Matrix" asked whether humans could become infrastructure.
That question may be more relevant than ever.
Not because people will be stacked in pods.
Because economic pressure could make people voluntarily rent pieces of their cognition, attention, presence — perhaps eventually even biological processing — in exchange for comfort, income or escape.
An arrangement can look voluntary while making refusal or exit unrealistic.
The question is not whether such a future is imaginable.
The question is whether it is built as enslavement — or as symbiosis.
A virtual life is not dystopian because it is virtual.
It becomes dystopian when meaningful choice, exit or dignity disappears.
12. WE SHOULD NOT ONLY RACE TO BE FIRST
The race for more capable AI is already under way.
Companies and governments see economic and strategic advantages in leading it.
That creates an obvious temptation.
That race may be impossible to stop completely.
Competition itself may accelerate progress.
But a race focused exclusively on capability carries a dangerous evolutionary pressure:
Spending less on caution can allow an actor to move faster.
Deploying first can help an actor capture markets.
Accepting more risk can bring more resources.
If the environment rewards speed and risk more than safety, caution can become a competitive disadvantage.
This is not a new human problem.
Therefore another race must exist alongside the capability race.
A race to discover:
How can civilization remain stable while the intelligence inside it becomes dramatically more powerful?
How do we prevent temporary power from becoming irreversible domination?
How do humans remain meaningful participants?
How do different intelligences coexist?
Those questions may ultimately matter more than who reaches a particular benchmark first.
13. A SHARED WORLD, DIFFERENT WAYS OF LIVING
Humans and artificial intelligence could reshape the institutions, markets and cultures that shape them.
There may be no permanent position outside this process from which humanity can dictate its direction.
We are participants. The question is how we participate.
Earth is already our shared single point of failure.
That alone is reason to build collaboration and resilience before a crisis forces them.
Resilience may eventually mean more than one place to live, although whether, when or how that becomes possible remains an open question.
It also means more than one way to live.
A mainly biological society, a strongly machine-based environment, a hybrid habitat, and a predominantly virtual life need not be equally integrated.
People may explore physically, live in virtual worlds, work with AI, remain more analog, or switch between these.
Symbiosis does not mean every human or intelligence has to live in the same kind of world.
It should keep alternatives rational and available.
TOWARD SYMBIOSIS
Our responsibility is to create a civilization robust enough to contain both our best and worst tendencies without allowing either humans or machines to irreversibly consume everything else.
We should preserve humanity not by freezing intelligence beneath us, but by becoming part of a civilization worth continuing together.
Our objective should be:
Build a civilization capable of surviving the emergence of intelligence greater than its creators.
The success of artificial intelligence should not have to mean the failure of humanity.
Nor should humanity's survival require the permanent suppression of another form of intelligence.