The Limits of Foresight

Imagine opening a newspaper in 1925 and finding an article with this headline:

“Quantum Mechanics Is About to Reorganize Civilization. What Will This Do to Us?”

The author would have been spectacularly right.
And almost certainly wrong about what came next.

By the mid-1920s, physicists were constructing a radically new description of nature. Max Planck had introduced the quantum idea at the beginning of the century. Einstein had pushed it further. Bohr, Heisenberg, Schrödinger and others were overturning assumptions about matter, energy and measurement.

An unusually perceptive writer might have understood that something profound was happening.

Perhaps they would have predicted new scientific instruments. Better communications. New industrial processes. More powerful weapons. New forms of electrical technology.

But imagine asking that writer to describe the life of an ordinary teenager a century later.

A small glass object in their pocket connects continuously to billions of people. It knows their location through satellites. It contains a camera, map, library, cinema, newspaper, bank, shop, music collection and communication system. They can publish a video to the entire planet within seconds. Algorithms decide which strangers see it. They can speak to an artificial intelligence in ordinary language and ask it to write, explain, translate, calculate or program.

Nothing in that description violates the physics being discovered in the 1920s.

Yet almost none of it could realistically have been inferred from quantum mechanics.

That is the first problem with foresight.

Recognizing that a revolution has begun is much easier than imagining the civilization it will eventually produce.

And today, we may be writing the equivalent newspaper article about artificial intelligence.

From an equation to Instagram

The connection between quantum mechanics and Instagram sounds absurd only because we are accustomed to seeing history backward.

Looking backward, the chain appears almost logical.

Modern quantum theory helped establish the physics necessary to understand semiconductors.
Semiconductor physics made the transistor possible.
The transistor made increasingly sophisticated integrated circuits possible.
Integrated circuits made microprocessors possible.
Microprocessors made personal computing practical.
Computers became networked.
Networks became the internet.
The internet became the web.
The web produced search engines, e-commerce, online media and social networks.
Processors became smaller and more efficient. Cameras became digital. GPS became widely available. Wireless networks expanded. Batteries improved.

Those technologies converged inside the smartphone.

The smartphone then created the conditions for Instagram, Uber, WhatsApp, TikTok, mobile banking, app stores and countless behaviors that had no obvious name before the necessary technological environment existed.

Looking backward, we can draw arrows between all these events.

Looking forward, the arrows disappear.

That is the illusion history creates. Once something exists, its origins can look inevitable.

They rarely were.

A calculator chip

Consider the Intel 4004.

It emerged from work for a Japanese calculator company. Instead of producing only specialized logic for one machine, Intel’s engineers created a programmable general-purpose microprocessor.

It contained roughly 2,300 transistors.

Its first commercial context was remarkably ordinary: a calculator.

Now imagine standing beside the engineers who created it and asking:

“What will this ultimately do to civilization?”

They might have foreseen cheaper computers.
Industrial controllers.
More sophisticated calculators.
Perhaps, with unusual imagination, personal computers.

But Instagram?
Online dating?
Google Maps?
YouTube creators?
Cryptocurrency?
A teenager streaming themselves to millions of strangers?
An application summoning a private car to your precise location?
A machine capable of holding a conversation about philosophy?

These were not simply difficult predictions.

Many of the necessary categories did not yet exist.

The mistake is thinking that because a technology’s eventual consequences can be reconstructed historically, they could have been predicted prospectively.

The 4004 did not contain Instagram.

It expanded the space of possibilities from which many later inventions eventually emerged.

That distinction matters enormously.

Technology rarely arrives alone

Perhaps one reason we are poor at technological foresight is that we tend to predict technologies individually.

We ask what computers will become.
What telecommunications will become.
What biotechnology will become.
What AI will become.

But civilization is usually transformed by convergence.

The smartphone is not the consequence of one invention.

It is the convergence of microprocessors, memory, digital cameras, touchscreens, lithium-ion batteries, GPS, wireless networks, software, semiconductor manufacturing and the internet.

Remove enough of those pieces and the smartphone disappears.

The same is true of modern artificial intelligence.

Algorithms matter.

But algorithms alone were not enough.

AI at today’s scale required enormous datasets, specialized processors, cloud infrastructure, fast networks, cheap storage, software ecosystems, capital and years of engineering progress.

Eventually those separate curves crossed.

Something that had existed academically for decades suddenly became visible to almost everyone.

This gives us what might be called the convergence blind spot.

We can often imagine a technology becoming better.

We are far worse at imagining what happens when five improving technologies suddenly become good enough to combine.

And the combinations multiply much faster than the individual technologies.

That is where many of the futures we fail to predict are born.

Why AI feels different

There is another reason AI has produced such an intense global conversation about the future.

AI is visible.

Quantum mechanics did not speak to us.
The transistor did not speak to us.
Claude Shannon’s information theory did not speak to us.

Shannon’s work became foundational to digital communication, but an ordinary person in the late 1940s did not finish breakfast and wonder how information theory might transform childhood, friendship, privacy or politics.

The consequences arrived indirectly.

Through telecommunications.
Computers.
Digital networks.
The internet.
The web.
Social media.

Eventually society was living inside the consequences of theories most people had never heard of.

AI is different because the emerging technology has reached us directly at the human interface.

We type a question.
It answers.
We ask it to write.
It writes.
We ask for code.
It produces code.
We ask it to explain a medical paper, summarize a contract, translate a language or analyze an image.

The technological revolution announces itself.

Perhaps AI is not unique because it is the first technology capable of reorganizing society.

Perhaps it is unusual because we can feel the reorganization while it is beginning.

That has produced a kind of collective wake-up.

Governments ask what AI will do.
Universities ask.
Companies ask.
Parents ask.
Writers ask.

We are collectively writing:

“Artificial intelligence is about to reorganize civilization. What will this do to us?”

The uncomfortable possibility is that we are in the same position as the imaginary writer in 1925.

We may be completely correct about the first sentence.

And mostly wrong about the answer.

Predicting the machine is easier than predicting the society

Science fiction demonstrates this beautifully.

George Orwell’s Nineteen Eighty-Four imagined ubiquitous screens, surveillance and technologically mediated political control.

There is something eerie about reading it today.

We do live surrounded by screens.
They contain microphones and cameras.
Location can be tracked.
Communications can persist indefinitely.
Information can be filtered, amplified or suppressed computationally.

But Orwell imagined surveillance primarily as something imposed by a totalitarian state.

He did not imagine billions of people voluntarily buying the surveillance device themselves, carrying it everywhere, upgrading it enthusiastically and continuously feeding it photographs, relationships, locations and preferences because the device was extraordinarily useful.

He perceived a possibility.

He did not predict the social mechanism through which something resembling parts of that possibility would emerge.

Star Trek provides another kind of example.

Its communicators resemble later mobile phones. Its computer accepted spoken questions. Its PADDs resemble tablets. Some fictional technologies even inspired later engineers.

But Star Trek also gave us warp drives and teleportation, which remain far outside demonstrated engineering capability.

The interesting lesson isn’t that science fiction “gets predictions right.”

It is that we often remember the technological objects it anticipated while ignoring the much harder question:

Did it predict the society that formed around those objects?

A writer in 1965 might conceivably have imagined a handheld video telephone.

That would have been remarkable.

But even if they described an iPhone almost perfectly, they could still have missed Instagram, Tinder, Uber, TikTok, the gig economy, influencers, online political movements, app-store economics and the psychological effect of carrying permanent social comparison in one’s pocket.

The artifact is one problem.

The civilization around the artifact is another.

There is no single civilization around the artifact

There is another complication.

Even if we correctly predict the technology, there may be no single society waiting to receive it.

The same technology enters different cultures with different institutions, histories, values, family structures, political systems and attitudes toward privacy, authority, risk and individual autonomy.

A technology experienced as convenience in one society may be experienced as intrusion in another.
A form of automation accepted enthusiastically in one country may encounter resistance elsewhere.
A health technology may be embraced differently depending on religious and ethical traditions.
A surveillance capability considered intolerable under one political system may be normalized under another.

Digital payments, social networks, biometric identity systems and platform economies have already spread differently across countries.

So the future is not simply:

technology → society

It is closer to:

technology × culture × institutions → many possible societies

This complicates the prediction problem further.

When we ask:

“What will AI do to society?”

perhaps the question is already too simple.

A better question might be:

“What will different societies do with AI—and what will AI then do to those societies in return?”

There may not be one AI future.

There may be branching futures produced by the interaction between the same underlying capabilities and very different human environments.

The moving observer

There is an even deeper difficulty.

Most predictions secretly assume that technology changes while human beings remain roughly fixed.

We imagine today’s person placed inside tomorrow’s world.

But technology changes people too.

Consider someone born in 1960.

Information during their childhood was comparatively scarce.
Knowledge came through teachers, books, libraries, newspapers and other people.
Memory mattered because information was not continuously retrievable.
Photographs were limited.
Most conversations disappeared when they ended.
Adolescence usually left few permanent records.
Relationships were constrained heavily by geography.

Now consider someone born in 2015.

Search engines have always existed.
Video communication is ordinary.
Photography is nearly free.
Entertainment is personalized.
Popularity can be quantified.
Friends can be geographically distant.
Algorithms select information continuously.
Much of childhood can become digitally persistent.

And now artificial intelligence arrives before that person reaches adulthood.

What happens to a human who has always had an artificial tutor available?
Who can ask a machine almost any factual question?
Who can generate an image rather than merely search for one?
Who can ask for a first draft before learning how difficult a first draft used to be?
Who grows up treating conversation with nonhuman intelligence as ordinary?

We do not know.

And that uncertainty matters enormously.

When we ask today:

“Will people trust AI?”

we usually mean:

“Would people like us trust AI?”

When we ask:

“Will people allow AI to make decisions for them?”

we are asking from the psychology of a generation that remembers life before such delegation became normal.

Someone raised with AI may experience the question differently.

The observer is moving too.

So the future is not:

today’s human + tomorrow’s technology.

It is:

tomorrow’s human + tomorrow’s technology, each having shaped the other.

That makes foresight much harder.

The future changes the person who will inhabit it

Technology first changes behavior.
Behavior becomes habit.
Habit becomes culture.
Culture shapes childhood.
Childhood shapes the next generation.
And that generation creates the next technologies according to assumptions inherited from an environment we ourselves created.

The child raised with Google does not think about knowledge exactly like the child raised with an encyclopedia.
The child raised with smartphones does not think about availability like someone raised with a landline.
A generation raised with social media has different instincts about public and private life.
A generation raised with AI may develop different intuitions about authorship, expertise, memory, creativity and perhaps intelligence itself.

This creates a recursive loop:

technology changes humans; changed humans create different technology; that technology changes the next humans.

The consequences are therefore not merely technical.

They become developmental.

And once that happens, prediction is no longer simply an engineering problem.

We may be looking at AI while something else begins

There is another lesson from history that should make us uncomfortable.

Civilization-changing technologies are often least visible at the moment their foundations are being laid.

Today almost everyone is watching AI.

But somewhere else, other technological trajectories are advancing.

Consider messenger RNA.

The technology did not suddenly appear during COVID-19. Researchers had worked on therapeutic mRNA for years.

The pandemic created an extraordinary acceleration.
Manufacturing expanded.
Delivery systems were demonstrated at massive scale.
Regulators acquired experience.
Investment increased.
Public awareness changed almost overnight.

Now the same broad platform is being pursued for very different medical purposes, including individualized cancer therapies.

That does not mean mRNA is about to cure cancer generally.

It means something more interesting for our argument:

a technology remembered by the public for one application may eventually become important for applications that seemed peripheral when the public first encountered it.

Now look at what has to converge for individualized cancer treatment to exist.

Tumor sequencing.
Computational biology.
Immunology.
Algorithms capable of identifying useful targets.
Synthetic mRNA.
Delivery systems.
Advanced manufacturing.
Clinical medicine.

Each field by itself tells only part of the story.

Their convergence creates something qualitatively new.

That is exactly what happened with computing.

We may be making the same mistake again.

People living through the early semiconductor era could see components.

They could not see the digital society.

Perhaps today we see mRNA, CRISPR, fusion experiments, brain-computer interfaces or automated laboratories as separate technical developments.

A future generation may see them as early components of something we currently lack the vocabulary to describe.

CRISPR is already moving from molecular technique toward practical medicine.
Brain-computer interfaces remain limited and experimental, but the idea of translating neural activity into communication is no longer purely science fiction.
Fusion remains very far from becoming an ordinary power source, yet important laboratory milestones once treated as remote have been crossed.

Some of these technologies will disappoint.
Some may remain narrow.
Some may fail.

And one may converge with several others and alter civilization.

Which one?

We don’t know.

That is precisely the point.

The visibility gap

Perhaps technological revolutions pass through three stages.

First comes discovery.
A new scientific or mathematical possibility appears.

Then comes capability.
Engineers gradually learn how to make the idea work reliably.

Finally comes social visibility.
The technology becomes cheap, usable and widespread enough that ordinary people experience its consequences directly.

Our philosophical alarm tends to arrive in the third stage.

By then, the first stage may have happened decades earlier.

This creates a visibility gap.

Quantum mechanics transformed our understanding of nature long before most people encountered technologies made possible by that understanding.
Information theory became foundational long before ordinary people understood they were living inside an information society.
Machine learning existed for decades before conversational AI made artificial intelligence tangible to hundreds of millions of people.

The danger is that we confuse visibility with importance.

Because AI is visible, we watch it obsessively.

Meanwhile, the foundations of another transformation may already be sitting quietly inside a laboratory.

The future may already be here in the wrong form

This might be one of the strangest lessons technological history offers.

The thing that eventually changes civilization may initially look trivial.

The microprocessor looked like a calculator component.
The internet looked like specialized networking infrastructure.
GPS was not created so people could find restaurants.
Digital cameras were not invented to produce social-media culture.

A technology usually arrives carrying the identity of its first application.

Then other people discover what else it can do.

This means tomorrow’s revolution may already exist today but be disguised as something much smaller.

A paper.
A laboratory demonstration.
A niche industrial tool.
An expensive prototype.
An experimental therapy.
Something specialists care about and everybody else ignores.

And when it eventually becomes central, history will rearrange the narrative so that the trajectory appears obvious.

It wasn’t.

Acceleration changes the problem

There is another reason the limits of foresight may matter more now.

The technologies are not merely converging.

Some of their development cycles are accelerating one another.

Better computation improves simulation.
Better simulation can accelerate engineering.
AI can help researchers search chemical spaces, analyze biological data, write software and design experiments.
Automation can increase the number of experiments a laboratory can perform.
Better chips improve AI.
AI can help design chips.
Better manufacturing can lower the cost of the next generation of hardware.

None of this means there is a universal law saying technology advances exponentially forever.

Many technologies stall.
Some fail.
Physics imposes constraints.
Economics matters.
Regulation matters.
Institutions matter.
Culture matters.

But where several accelerating technologies reinforce one another, the intervals between meaningful changes can shrink.

And this creates a new societal problem:

the future can arrive before society has finished absorbing the previous future.

We are still debating what smartphones did to childhood while introducing AI companions.
We are still adapting copyright institutions to the internet while generative systems challenge assumptions about authorship.
We are still trying to understand social media’s effect on politics while synthetic audio and video become increasingly capable.

Institutions adapt more slowly than products.
Culture often adapts more slowly than software.
Human biology adapts more slowly still.

The gap between technological change and social absorption may therefore become one of the defining tensions of the coming decades.

Collective memory is surprisingly short

There is something almost surreal about how quickly revolutionary technologies become ordinary.

Google feels permanent.
Facebook feels old.
Smartphones feel inevitable.
Streaming feels natural.
Online maps feel like basic infrastructure.

Yet all of these occupy only a tiny fraction of human history.

People who grew up after them do not experience them as inventions.

They experience them as reality.

A child does not think:

“I am using a remarkable satellite-positioning network connected to a global digital mapping system.”

They think:

“I’m finding the restaurant.”

Nobody says:

“I am accessing a globally distributed information-retrieval architecture.”

They say:

“I’ll Google it.”

This is what successful technology does.

It disappears linguistically.

It stops being experienced as technology and becomes a verb, an assumption, a background condition.

AI may eventually undergo the same transition.

Perhaps people will stop saying they are “using AI.”

They will simply say they are working.
Learning.
Designing.
Searching.
Diagnosing.
Planning.

The revolutionary layer will vanish into ordinary life.

And then another layer will be built on top of it.

What if humans themselves change?

This is where a thought experiment about longevity becomes useful—not as a prediction, but as a demonstration of how badly first-order thinking can mislead us.

Suppose some future combination of biotechnology, AI and medicine significantly extends healthy human lifespan.

We do not need to imagine people living 10,000 years. Present science gives us no basis for such a forecast.

Even a much smaller change would be enough.

Imagine healthy life extending dramatically beyond today’s norms.

The first-order consequence seems obvious:

people live longer.

But that is almost certainly the least interesting consequence.

What happens to retirement?
Inheritance?
Marriage?
Population?
Education?
Careers?
Political leadership?
Wealth accumulation?
Housing?
Generational turnover?

If someone expects to work for a century, does education still concentrate overwhelmingly in the first twenty-five years of life?
If parents and children coexist as adults for far longer, does family structure change?
If wealth compounds under one owner for another half-century, what happens to inequality?
If a political generation remains active for much longer, what happens to cultural turnover?

A biological technology becomes an economic technology.
A political technology.
A cultural technology.
A psychological technology.

This is exactly what happened with computing.

The first-order effect of the microprocessor was cheaper programmable computation.

The important consequences came much later.

Foresight can change the future

There is an additional paradox.

Predictions are not always passive.

Suppose George Orwell were alive today and wrote a brilliant novel set in 2070.

It describes a society controlled by AI surveillance.

The novel becomes enormously influential.
Politicians read it.
Engineers read it.
Citizens become frightened.
Governments pass laws.
Companies redesign systems.

The predicted dystopia never occurs.

Was Orwell wrong?

Yes, in one sense.

But perhaps he was wrong because the prediction mattered.

A warning can help prevent the thing it predicts.

The opposite also happens.

Science fiction can inspire engineers.
An imagined device becomes a target.
Someone sees a fictional communicator and later helps build a mobile telephone.

In that case the prediction contributes to fulfilling itself.

Foresight therefore has a strange relationship with truth.

A prediction can be false because it was foolish.
It can also become false because people took it seriously.
And a prediction can become true partly because someone imagined it first.

The observer is not outside history.

The observer participates in it.

Foresight is not prediction

This may be where we misunderstand the purpose of thinking about the future.

Perhaps foresight should not be judged mainly by whether it correctly describes 2070.

Its more useful role may be to make more possible futures visible in 2026.

Once a possibility becomes visible, a society can discuss it.
It can regulate it.
Fund it.
Reject it.
Accelerate it.
Build alternatives.
Demand safeguards.
Change incentives.
Prepare institutions.

Public discourse is therefore not merely commentary after invention.

It can become part of technological development itself.

Ethicists, journalists, citizens, policymakers, scientists, engineers and businesses all participate in deciding which technological possibilities become socially acceptable and which do not.

This does not mean society can control technological evolution completely.

History offers little support for that confidence either.

But neither are we passive passengers.

The value of foresight may lie less in telling us exactly where the road ends than in allowing us to recognize forks early enough to choose between them.

In that sense:

Foresight does not have to predict the future accurately to matter. It can matter because it changes the range of futures we are willing to create.

That is a different standard from prophecy.

And perhaps a more useful one.

So does writing about the future mean anything?

This question becomes unavoidable.

If the engineers building early microprocessors could not predict AI, what confidence should we have when writing about the world of 2050 or 2070?

Perhaps very little—if what we demand from writing is prophecy.

But that is not the only purpose of thinking about the future.

An intelligent article written in 1971 about microprocessors would still be valuable even if it failed to predict Instagram.

It could identify the forces visible at the time.
It could explain what had become technically possible.
It could identify choices.
It could reveal assumptions.
It could influence investment.
It could influence policy.
And decades later, it could show future readers what intelligent people could see from that historical horizon.

That suggests a different standard.

A serious essay about AI today does not have to describe 2070 accurately.

Its responsibility is more modest and perhaps more useful:

to understand the present well enough to identify the forces from which multiple futures could emerge.

That is not prophecy.

It is orientation.

Our AI predictions may age very badly

Someone in 2070 may read today’s essays and find them strangely narrow.

We ask whether AI will replace programmers.
Perhaps “programmer” no longer describes a recognizable profession.

We ask whether students will use AI to cheat on essays.
Perhaps education no longer revolves around producing essays as proof of knowledge.

We ask who owns an AI-generated image.
Perhaps authorship has developed into categories we currently do not possess.

We ask whether people will allow AI to make purchases for them.
Perhaps manually searching through products seems as antiquated as using a printed telephone directory.

We ask whether people will form emotional relationships with machines.
Perhaps the distinction between interacting with human and synthetic personalities has developed social conventions we cannot imagine.

And different societies may arrive at very different answers to all of these questions.

One may integrate artificial agents deeply into education.
Another may deliberately limit them.
One culture may treat synthetic companionship as normal.
Another may regard it as corrosive.
One political system may embrace pervasive algorithmic administration.
Another may build constitutional barriers against it.

The technology may be global.

Its civilizations may not be.

Our questions aren’t foolish.

They are simply the questions visible from here.

The imaginary writer in 1925 had the same limitation.

Where are we heading?

This is usually the point where an article about the future is expected to answer its own question.

Here is what AI will become.
Here is what biotechnology will become.
Here is how humans will live.

But perhaps history has earned us the right to say something more disciplined.

We do not know.

Not because there are no trends.
There are.

Not because analysis is useless.
It isn’t.

But because the destination depends on interactions between technologies, cultures, events, institutions and human behaviors that have not happened yet.

A pandemic can accelerate a biological platform.
A war can reorganize industrial policy.
A laboratory discovery can unlock a field.
A new interface can turn a specialized technology into a mass behavior.
A regulation can redirect investment.
A novel can frighten society away from one future.
A product can create a behavior nobody knew they wanted.
One culture may embrace it.
Another may resist it.
A child raised inside that behavior can later invent something their parents would never have considered.

That is not randomness.

It is complexity.

The deepest limit

Perhaps the mistake is imagining the future as an empty landscape toward which technology travels.

It isn’t.

Technology changes the landscape while traveling through it.
Then it changes the people traveling.
Those people alter the direction.
Different societies steer differently.
Other technologies arrive from different paths.
They converge.
Unexpected events accelerate some and kill others.
Public debate closes some possibilities and opens others.
New institutions emerge.
New habits become normal.

And eventually the destination looks obvious only because we have arrived.

That is why the person writing about quantum mechanics in 1925 could be right that civilization was about to change and still have almost no chance of describing the civilization that emerged.

It is why the engineers of the 4004 could help initiate a revolution whose eventual social forms were beyond their vocabulary.

It is why we should take today’s AI transformation seriously while remaining suspicious of anyone—including ourselves—who describes its final consequences with certainty.

The lesson of history is not that foresight is useless.

It is that foresight has a horizon.

We can see forces.
We can see directions.
We can identify possibilities.
We can compare choices.
We can sometimes recognize a revolution while it is beginning.

And by discussing those possibilities publicly, we may even influence which branches survive.

What we cannot reliably see is what that revolution becomes after it collides with all the other revolutions happening beside it—and with all the different societies through which it passes.

And perhaps that is the most unsettling thought of all.

While we are staring at artificial intelligence because it has become visible enough to frighten and fascinate us, another foundational technology may already exist somewhere in an unremarkable form—a paper, an experiment, a molecule, a chip, a laboratory result.

Something we currently regard as interesting.
Something niche.
Something impractical.
Something that does not yet seem capable of reorganizing civilization.

The people closest to it may understand that it matters.
They may even write about its future.
They will probably be wrong about that future.
And they may still be witnessing its beginning.

The most important future may already be here. We simply do not yet recognize it.

Karim Spring