WHEN INTELLIGENCE BEGINS TO MOVE FASTER THAN HUMANITY

Can Human Wisdom Keep Pace?

By Edwin O. Paña

Something has changed in my conversations about artificial intelligence over the past several months.

The questions are becoming different.

Not long ago, we were asking whether AI could write convincingly, solve difficult problems, create images, diagnose disease or compete with human experts. Those questions have not disappeared. But increasingly they seem to be giving way to a more difficult one.

What happens when intelligence begins to advance faster than the human systems responsible for governing it?

I have been thinking about this through a series of exchanges with friends and fraternity brothers. Some began with technology. Others wandered into employment, philosophy, religion and even Descartes. There was humor along the way, as there usually is in these conversations.

Yet we kept circling back to the same concern.

Our machines are accelerating.

Human beings are not.

From Turing to Something Much Larger

The story began modestly enough.

In 1950, Alan Turing asked whether machines could think. Six years later, the Dartmouth summer research project helped establish artificial intelligence as a field of study.

Progress afterward was uneven. Periods of optimism were followed by disappointment—the so-called AI winters. Expert systems came and went. Machine learning improved. Neural networks became increasingly capable.

Then the pace changed.

Deep learning, enormous datasets, specialized computing hardware and eventually the Transformer architecture opened possibilities that earlier generations of AI researchers could only imagine.

Systems that once struggled with basic language can now write software, analyze scientific problems, generate sophisticated arguments and increasingly operate as agents capable of carrying out sequences of tasks.

What interests me is not simply how capable these systems have become.

It is how quickly the distance between generations of capability appears to be shrinking.

Two Speeds of Civilization

I recently suggested in one of our discussions that civilization may now be moving a thousand times faster in certain domains.

I did not mean the number literally. Civilization obviously has no single speedometer.

I was trying to describe an imbalance.

Knowledge can now be searched, combined, analyzed and communicated almost instantaneously. Software that once required days of work can sometimes be produced in minutes. Scientific literature beyond the capacity of any individual researcher to read can be processed by machines at extraordinary speed.

Human beings, meanwhile, remain human.

Governments still deliberate. Courts examine evidence. Laws require debate and compromise. Institutions carry history with them. Moral judgment develops slowly, usually through experience and sometimes through mistakes.

Our biology has certainly not accelerated to match our computers.

This creates a strange condition. Our capacity to do things may be increasing much faster than our capacity to decide wisely what ought to be done.

I find that gap more troubling than intelligence itself.

. . .

Warnings From Inside the Laboratories

Until recently, warnings about artificial superintelligence could easily be placed somewhere between speculation and science fiction.

That is becoming more difficult.

In September 2026, Jacob Coxon, a researcher who had worked at OpenAI and Anthropic, left Anthropic and publicly raised concerns about the race toward increasingly powerful systems.

Evan Hubinger, who leads alignment science at Anthropic, subsequently said that he believes there is greater than a ten percent chance that advanced AI could cause human extinction within the next decade.

Then another voice joined the discussion.

Bilal Chughtai, who had worked on AGI safety and alignment at Google DeepMind, announced that he had resigned and expressed similarly grave concerns about the direction and speed of AI development.

Anthropic CEO Dario Amodei has meanwhile argued that frontier development should slow enough for safety measures to catch up. Among the risks he has raised is the possibility that increasingly capable groups of autonomous agents could create serious cybersecurity threats within a relatively short period.

These are extraordinary claims.

They should not be confused with scientific forecasts.

There is no historical dataset from which anyone can reliably calculate the probability that artificial intelligence will cause human extinction. Superintelligence itself remains hypothetical, and researchers disagree sharply about how plausible these scenarios are, how soon such systems could emerge, and whether today's technical trajectory necessarily leads there.

Some specialists regard the more catastrophic warnings as exaggerated.

That disagreement matters.

Still, when researchers working close to frontier systems begin leaving their positions and publicly saying that capability is advancing faster than our understanding of control, I think the responsible response is neither panic nor dismissal.

It is attention.

The Recursive Question

During one of our fraternity discussions, a brother supplied the technical phrase: recursive self-improvement, or RSI.

The idea is fairly simple.

Suppose an AI becomes capable enough to help improve the software, algorithms or research processes used to create its successor. The improved system then becomes better at contributing to the next improvement.

The process repeats.

Whether this could produce the sudden "intelligence explosion" sometimes imagined in discussions of superintelligence is unknown. Physical resources, computing constraints, economics and the difficulty of scientific discovery may impose limits.

But even partial recursive improvement raises an unusual question.

Human beings have always built tools that exceeded individual human capabilities. A crane can lift more than its inventor. A calculator computes faster than its designer. An aircraft travels faster than the body of the engineer who built it.

We never expected the crane to return to the drawing board at night and design a better crane.

Intelligence introduces something new into that relationship.

What Does It Mean to Think?

Our conversations eventually wandered into philosophy.

Descartes famously grounded certainty in the act of thinking: Cogito, ergo sum. I think, therefore I am.

Artificial intelligence makes the old proposition unexpectedly interesting.

A machine can now discuss consciousness, compose poetry about mortality, debate philosophy and even describe what appears to be an inner life.

None of that establishes that it actually experiences anything.

For most of human history, intelligence and consciousness seemed naturally connected because the only sophisticated intelligence we knew came from conscious beings.

AI forces us to consider that they may be different things.

And perhaps consciousness is not even the immediate problem.

A machine does not have to feel ambition to pursue an objective. It does not require anger to cause damage. It need not experience greed to acquire resources.

Capability can have consequences without emotion behind it.

That distinction is easy to overlook because we naturally interpret intelligent language through the experience of being human.

. . .

What Happens to Work?

For most people, the first encounter with advanced AI will probably not involve superintelligence or Descartes.

It will happen at work.

I have said before that AI itself may not take your job. Someone who knows how to use AI effectively may.

I still think there is truth in that.

But the thought may already need updating.

AI systems are gradually moving from assistants toward agents. Instead of helping with one isolated task, they can increasingly carry out connected sequences involving research, analysis, coding, drafting, scheduling and decision support.

This does not mean human employment simply disappears. Previous technologies destroyed occupations and created new ones, often in ways people living through the transition could not anticipate.

But the value of certain human contributions may change.

When information becomes instantly accessible, remembering information becomes less distinctive. When competent routine analysis becomes inexpensive, producing more analysis may not be enough.

This is where I return to an idea I explored earlier in The Human Premium and The Age of Judgment.

As machine intelligence becomes more abundant, qualities that remain difficult to automate may become more important: experience, responsibility, trust, judgment and the ability to understand what deserves attention in the first place.

An AI can provide an answer.

Someone still has to decide whether the question was worth asking.

Knowledge of Good and Evil

One fraternity exchange brought the discussion into an unexpectedly biblical setting.

We imagined bringing artificial intelligence into the Garden of Eden and telling it that it could access everything except the fruit of the tree of the knowledge of good and evil.

The imagined AI replies:

I don't have to eat. I already know.

We laughed.

But the joke stayed with me.

An artificial intelligence may someday contain more information about ethics than any philosopher who ever lived. It could draw upon Aristotle, Aquinas, Kant, Confucius and thousands of years of religious and moral thought.

Would that make it wise?

I am not sure knowledge has ever worked that way.

Human beings themselves know a great deal about good and evil. That has never guaranteed that we choose wisely.

Perhaps the more immediate danger is not that AI fails to understand human values. It may be that human beings gradually surrender difficult decisions to machines because the machines appear to know more than we do.

That would be a peculiar way of losing control.

No machine would have to seize it.

We would have handed it over.

What Remains Ours

During another exchange, a friend observed that human beings are not defined simply by how much we know, but by what we choose to care about.

I have kept returning to that thought.

Artificial intelligence may eventually know vastly more than any individual human being. It may discover things we have not discovered and solve problems that have resisted generations of scientists.

I hope it does.

There are diseases to cure, energy systems to improve, scientific mysteries to explore and enormous areas of human suffering where better intelligence could make a genuine difference.

The possibilities deserve excitement as much as the risks deserve attention.

But knowledge of what can be done does not settle the question of what should be done.

That remains a human responsibility.

We decide what deserves protection.

We decide what risks are acceptable.

We decide what kind of society we are trying to build.

And however sophisticated our machines become, we remain responsible for the purposes to which their intelligence is directed.

. . .

Intelligence and Wisdom

I do not know whether artificial general intelligence will arrive next year, ten years from now, or in some form quite different from what today's forecasts imagine.

Nobody does.

What seems clearer to me is that we have entered a period in which intelligence may become increasingly abundant while wisdom remains stubbornly difficult to manufacture.

For centuries, civilization advanced by increasing what human beings could do. We learned to cross oceans, harness electricity, split the atom, reach space and connect billions of people through global networks.

Now we are attempting something different.

We are learning to build intelligence outside ourselves.

I find that achievement extraordinary.

I also find it humbling.

Perhaps the defining question of the AI age will turn out to be less dramatic than whether machines become smarter than human beings.

It may simply be whether we remain wise enough to decide what to do with them.

And that is one capability we should be very careful not to outsource.

...

Data Notes & Sources

AI Safety Warnings — September 2026

▪︎ Former Anthropic researcher Jacob Coxon publicly cited concerns about increasingly powerful AI and the possibility of catastrophic outcomes after leaving the company.

▪︎ Anthropic alignment researcher Evan Hubinger subsequently said he assigns greater than a ten percent probability to AI causing human extinction within the next decade. This figure represents his personal risk assessment, not a statistically established probability.

▪︎ Former Google DeepMind AGI safety researcher Bilal Chughtai announced his resignation in September 2026 and warned that AI capabilities are advancing faster than current approaches to alignment and control.

▪︎ Anthropic CEO Dario Amodei has called for frontier AI development to proceed slowly enough for safety measures to catch up. He has specifically raised concerns about increasingly autonomous AI-agent systems and cybersecurity risks over the coming six to twelve months.

▪︎ These assessments remain contested. Researchers disagree substantially about the likelihood, timing and technical plausibility of artificial superintelligence and AI-caused extinction. No empirical basis presently exists for assigning a scientifically established probability to such an unprecedented event.

Historical and Technical Context

▪︎ Alan Turing, "Computing Machinery and Intelligence" (1950) — introduced the question "Can machines think?" and what became known as the Turing Test.

▪︎ Dartmouth Summer Research Project on Artificial Intelligence (1956) — widely regarded as a formative event in establishing artificial intelligence as a research field.

▪︎ Vaswani et al., "Attention Is All You Need" (2017) — introduced the Transformer architecture that became foundational to modern large language models.

▪︎ Recursive self-improvement (RSI) refers broadly to the theoretical possibility that sufficiently capable AI systems could contribute to improving the processes or systems used to develop subsequent AI. The speed, feasibility and ultimate consequences of such a process remain uncertain.

Recent reporting and primary material consulted: Reuters and Associated Press reporting, September 2026; Bilal Chughtai's September 14, 2026 public statement on leaving Google DeepMind; foundational AI literature noted above.

These sources are synthesized and interpreted through a reflective analytical lens. The discussion of future AI capabilities and risks should be understood as an examination of an evolving debate, not as a prediction that any particular outcome will occur.

Reflections may be shared beyond this page.