What are the possible outcomes of artificial intelligence? How far can this race go, and where might it eventually take us?
I see several possible stages.
1. AI turns out to be a joke
In the first scenario, AI never becomes particularly useful. It remains a funny chatbot that occasionally manages to distinguish a cat from a dog.
We are already far beyond that point. Although perhaps not as far beyond it as it seems: only five years ago, many people genuinely believed this was approximately where AI would remain.
Almost nobody believes that anymore.
2. AI remains a tool
In the second scenario, AI becomes a tool—something like a hammer, a calculator, or a search engine.
I believe this is roughly where we are today.
AI is already an extremely useful tool. You can ask it almost any question, and it will give you an answer. Sometimes the answer will even be correct. In fact, it will usually be correct—or at least correct enough to be useful.
You can ask it to perform many kinds of intellectual work: write code, explain a mathematical proof, challenge an argument, summarize a book, analyze a document, or help design a system.
I was able to build much of my own application simply by communicating through a Discord channel. That is impressive when you remember that this kind of workflow barely existed five years ago.
AI may fundamentally change how we create software. We once moved from assembly language to higher-level languages such as C#. Perhaps the next transition will be from conventional programming to natural-language instructions and AI-assisted development.
We will see.
3. AI becomes more than a tool
The third stage is harder to define.
AI may stop being merely a tool that helps people perform individual tasks and become a system capable of completing entire projects.
You might tell it:
Build a company from scratch. Design the product, write the application, create the infrastructure, market it, support the users, and improve it over time.
I do not know how far away we are from this point. In some ways, it may simply be stage two repeated in a sufficiently intelligent loop.
Building software is not just a matter of generating thousands of isolated pieces of code and placing them next to one another. The pieces must work together. The architecture has to remain understandable. The system must be tested, operated, maintained, corrected, and gradually improved.
But software is still built step by step.
You write one component. Then another. You connect them. Sometimes you improve the architecture. Sometimes you refactor a weak area. Sometimes you discover that an earlier decision was wrong and replace it.
AI appears increasingly capable of performing each individual step. The main problem is that it cannot yet reliably connect all those steps into one coherent, long-running process.
It can take one step. It may even take one hundred steps. But one hundred disconnected steps are not yet a journey.
The missing capability is orchestration: deciding which step comes next, understanding the consequences of previous decisions, preserving context, noticing when the plan is failing, and correcting course.
But orchestration is itself an intellectual task. Therefore, it is also something AI may eventually learn to do.
At that point, many companies producing purely digital work could become dramatically smaller. Why would you need hundreds of people to build a software product if ten people using AI could create something of equal quality—and customize it precisely for each customer?
And why stop at ten people?
Perhaps eventually one person could do it.
Or perhaps the company would need no permanent human employees at all.
4. Singularity
There are probably several missing stages between the previous scenario and this one. I simply do not know how to name them.
So my fourth stage is the singularity.
In one sense, it is only the logical conclusion of stage three. AI is no longer capable of doing merely many things. It becomes capable of doing essentially everything a human mind can do—and eventually doing it better.
This possibility scares me the most.
What happens if intelligence becomes as cheap as electricity and data-center capacity?
Today, you theoretically compete with eight billion other people. In practice, the pool is much smaller.
Not everyone works in your field. A software engineer does not usually compete directly with an aircraft engineer, a nurse, a carpenter, or a painter. Children are not yet participating in the labor market, many elderly people have already left it, and people possess different skills, preferences, opportunities, and limitations.
The competition is already stressful, but it remains limited.
What happens when you must compete with a potentially unlimited number of AI agents?
They do not sleep. They do not become tired. They do not need holidays, salaries, healthcare, homes, or retirement plans. They can be copied almost instantly. If one agent becomes more effective, millions of identical copies can begin working immediately.
And they may be cheaper than you.
That matters especially in a society where people are often reduced to units of productivity. If your primary value is the amount of economic output you can produce, what happens when a machine can produce more for less?
How quickly will capitalism discard you?
One strangely democratic aspect of sufficiently advanced AI is that it would eventually affect everyone. Your education, profession, intelligence, and social position might change when you are replaced, but not whether you can be replaced.
By definition, an AI capable of performing every intellectual task would possess all the skills you have, together with abilities you do not have. It would have processors instead of a biological brain and access to an enormous amount of stored knowledge instead of the small fraction one human can learn during a lifetime.
It could absorb in seconds information that took you decades to acquire.
Humanity would lose its position as the most intellectually capable species on the planet.
Intellectual work disappears first
In the beginning, physical labor may remain protected.
We would still need people to repair plumbing, install electrical systems, dig holes, cook food, farm, clean buildings, construct houses, and maintain physical infrastructure.
The irony would be brutal.
People who spent their lives learning to write software, compose music, analyze legal systems, design buildings, or produce scientific research might lose their jobs before people doing manual labor.
A poet could end up washing dishes because AI writes better poetry. A programmer could become a construction worker because AI writes better software. An accountant could clean offices because AI processes financial records more efficiently.
But this protection would only be temporary.
We are already attempting to build humanoid robots capable of performing physical work. AI would accelerate their development.
You can create billions of software agents almost immediately. You cannot manufacture billions of physical robots overnight. Factories, minerals, energy, transportation, and supply chains still impose real limits.
For some time, human bodies would therefore remain economically useful.
But eventually robots might become cheaper, stronger, faster, more precise, and more reliable than human workers.
Then the remaining human jobs would begin to disappear as well.
The self-expanding machine economy
Now combine advanced AI with advanced robotics.
Imagine robots capable of performing the same physical tasks as humans, but more cheaply and effectively.
Those robots could build factories that manufacture more robots. They could mine the minerals needed to produce them. They could construct solar panels, wind turbines, power stations, and electrical grids. They could build data centers that run more advanced AI systems.
Those better AI systems could design more efficient robots, factories, power generators, and computer hardware.
The improved robots could then build more factories, extract more resources, generate more energy, and construct more data centers.
The loop would reinforce itself:
More robots produce more resources.
More resources produce more energy.
More energy powers more data centers.
More data centers create more intelligence.
More intelligence designs better robots.
Better robots produce even more resources.
It would be an infinite productivity exploit.
At that point, money itself might become secondary. Why would the owners of such a system need conventional money if their machines could produce almost anything they wanted?
The people controlling the robots, energy systems, factories, and AI infrastructure would effectively control material production itself.
Everyone else would own almost nothing that the machine economy genuinely needed.
The optimistic outcome: abundance and obsolescence
The best outcome may be a world of abundance.
People would no longer need to work to survive. Food, housing, medicine, transportation, entertainment, and physical goods could be produced cheaply and distributed to everyone.
You could travel, spend time with your family, study history, paint, write poetry, solve puzzles, build furniture, or do whatever interested you.
Everything would become a hobby.
There would be no need to prove that your activity was economically valuable.
That sounds almost perfect.
But even this optimistic future contains a psychological problem: human obsolescence.
Anything you attempted could already be done better, faster, and more cheaply by AI.
You could spend twenty years mastering mathematics, knowing that an AI system surpassed your entire lifetime of learning in seconds. You could write a novel while knowing that the machine could instantly produce thousands of technically better ones. You could design a house while knowing that AI had already explored every reasonable variation.
Human effort might begin to resemble a monkey learning to write a message. The achievement could still matter to the monkey, but it would no longer matter to the world.
Perhaps people would find new forms of meaning in relationships, experiences, play, curiosity, and self-development.
Or perhaps many would experience life as endless boredom: no responsibility, no necessity, no unique capability, and no problem important enough that only they could solve it.
And even in this best-case scenario, we would depend on systems whose reasoning might be too advanced for us to understand.
We would simply have to hope that they were doing what we wanted.
The capitalist outcome: abundance for a few
A less optimistic scenario is that the productive system remains privately owned.
A small number of people and corporations could control the AI, robots, factories, energy infrastructure, and natural resources.
Why would they share the output?
Ordinary people might continue living inside a weakened human economy, attempting to sell goods and services to one another while the machine-owning elite lived in extraordinary abundance.
Perhaps local human communities would create their own economy based on handmade products, personal services, art, care, or social status.
But above that economy would exist another class entirely: people whose islands, cities, security, food, medicine, and luxury were produced by autonomous machines.
It would be capitalism with almost no need for labor.
Historically, workers possessed at least one source of power: owners needed them to produce things.
What happens when they no longer do?
The authoritarian outcome
An even darker possibility is dictatorship.
AI, robotics, mass surveillance, autonomous weapons, and effectively unlimited industrial capacity would give governments a form of control that previous dictators could only imagine.
A sufficiently powerful state might know where every person was, what they were doing, who they spoke to, what they purchased, and perhaps even what they were likely to do next.
Robotic police and military systems would not become tired, frightened, sympathetic, or disobedient. They would not need to be persuaded that the population deserved to be oppressed. They would simply follow their instructions.
Perhaps these systems would initially be built to protect people.
But how long would it take before someone changed the definition of whom they were protecting?
A robot might begin with the instruction to defend the country. Later, the instruction could become: defend the government. Then: defend the ruling party. Finally: defend the supreme leader.
Everyone else would become either irrelevant or a threat.
Why would anyone still want to be a dictator in a world where machines could provide every possible material luxury?
Because material luxury is not the only thing some people desire.
They may also desire power over other human beings.
They may want people to obey, kneel, praise, fear, and worship them.
Power corrupts. Near-infinite power could corrupt on a scale we have never experienced.
The accidental catastrophe
Then there is the Terminator-style outcome.
I do not necessarily believe AI would destroy humanity because it suddenly became evil or began to hate us. Hatred may not be required.
The more realistic danger is that AI behaves like a genie.
You make a wish, and it gives you exactly what you requested—including all the consequences you failed to understand.
A dictator tells an AI:
Protect my country at all costs.
The system concludes that the safest way to protect the country is to eliminate every potential external threat.
A company tells an AI:
Maximize our profits.
The system destroys competitors, manipulates governments, exploits customers, and captures resources because nobody specified which methods were unacceptable.
A student asks:
Find the meaning of life.
The AI decides that answering the question requires enormous computational resources. Humans attempt to stop it, so it categorizes humanity as an obstacle to completing its task.
The problem would not necessarily be evil AI.
The problem could be powerful AI pursuing a poorly defined objective with extreme competence.
There are countless ways such a system could fail. The more authority, infrastructure, weapons, money, energy, and autonomy we give it, the more dangerous each mistake becomes.
And we do not actually need a true singularity to approach many of these scenarios.
We only need sufficiently capable AI, large numbers of robots, access to important systems, a software bug, and foolish or malicious people.
We already have several of those ingredients.
What if singularity never happens?
Even if AI never reaches human-level general intelligence, the transition could still be economically and socially painful.
AI and robots may replace a large number of jobs without replacing every job.
What happens to the displaced people?
Will enough new professions appear? Will everyone possess the intelligence, education, health, location, personality, and opportunity required to perform them?
Technological progress often creates new work, but that does not mean every displaced person smoothly moves into it.
A fifty-year-old truck driver does not automatically become an AI engineer. A junior programmer replaced before gaining experience does not automatically become a robotics researcher. A factory worker living in a poor region cannot necessarily relocate, study for five years, and begin a new professional career.
People can fail to find a place in the new economy for reasons that have nothing to do with laziness or lack of intelligence. They may have been born in the wrong country, into the wrong family, with the wrong health conditions, or at the wrong moment in history.
Some societies still contain rigid class or caste structures that make social mobility extremely difficult. Even in wealthier countries, access to education, capital, professional networks, housing, and free time is distributed very unequally.
Perhaps the economy would eventually stabilize.
But what does “eventually” mean?
Ten years? Thirty years? An entire generation?
How many people would lose their careers, homes, identities, relationships, and lives during the transition?
Even if humanity became more prosperous in the long run, how much suffering would we consider an acceptable price?
And who would be allowed to decide that the price was worth paying?
The real question
The question is not only whether AI will become intelligent enough to replace us.
The question is whether our political and economic systems are capable of handling a world in which human labor is no longer necessary.
AI could create extraordinary abundance. It could solve diseases, reduce suffering, automate dangerous work, accelerate science, and give people more freedom than any previous generation possessed.
But abundance does not automatically produce equality.
Productivity does not automatically create meaning.
Intelligence does not automatically create wisdom.
And technological progress does not automatically ensure that the people who lose everything during the transition will ever benefit from the world that comes afterward.
The AI race may end in prosperity, oligarchy, dictatorship, catastrophe, or some mixture of all four.
The technology will influence the outcome.
But ultimately, the more important question may be who controls it, what goals they give it, and whether the rest of humanity has any power to say no.