The word artificial does a lot of work.

Artificial flavor.

Artificial light.

Artificial flowers.

Artificial intelligence.

The word quietly tells us how to think before the conversation even begins:

There is the real thing.

And then there is the thing humans manufactured to imitate it.

But what if intelligence doesn't divide quite so neatly?

That is the rabbit hole opened by AI researcher Blaise Agüera y Arcas, Google’s CTO of Technology & Society and founder of its Paradigms of Intelligence research organization. His argument is not merely that computers can imitate intelligent behavior. He is asking whether biology and modern AI may reveal some of the same underlying principles of computation, prediction, cooperation and complexity.

Come on.

There is a door down here.

We often hear the brain compared to a computer.

Agüera y Arcas wants to remove the word compared.

His position is that nervous systems literally perform computation: they take information, transform it, generate signals, model circumstances and produce responses. He connects this with the idea that prediction is fundamental to what brains do and to what contemporary AI systems do.

That does not mean your brain resembles the laptop on your desk.

The hardware is radically different.

One is biological tissue containing billions of living neurons operating through electrochemical processes.

The other is silicon, circuits and mathematical operations.

But here's our first rabbit-hole distinction:

A bird flies with feathers.

An airplane flies with aluminum, turbines and control surfaces.

Flight is still flight.

So the deeper question becomes:

If intelligence is a process rather than a particular substance, must it be biological to be genuine?

Now the tunnel starts bending.

Agüera y Arcas defines intelligence roughly around an organism or system's ability to predict and influence what happens next. His 2025 book, What Is Intelligence?, develops the argument that prediction may connect brains, biological life and artificial systems more deeply than we ordinarily assume.

Think about how much of ordinary intelligence works this way.

You reach for a falling glass because you predict where it will be.

You finish another person's sentence because you predict the missing words.

You avoid a dangerous driver because you predict his trajectory.

You tell a joke because you predict someone else's expectations.

You plant seeds because you are modeling a future that does not yet exist.

Modern language models do something far narrower but strangely resonant:

They predict what should come next.

That alone doesn't make a language model a human mind.

But it does make the boundary considerably more interesting.

Because perhaps prediction is not a cheap imitation of intelligence.

Perhaps prediction is one of intelligence's oldest building blocks.

Now Hatta removes another floorboard.

Brains are relatively recent arrivals in the history of life.

Life existed long before nervous systems.

Long before animals.

Long before anything capable of wondering whether its intelligence was artificial.

Agüera y Arcas therefore pushes computation deeper into biology. His argument is that living systems have always had to process information: sensing conditions, maintaining themselves, reproducing, reacting, adapting and eventually predicting increasingly complicated environments.

And that leads to one of his most provocative propositions:

Life may have been computational from the beginning.

Not because prehistoric cells contained tiny laptops.

Because computation, in this broader sense, means transforming information in ways that affect what happens next.

DNA stores information.

Cells regulate processes.

Organisms sense environments.

Nervous systems coordinate increasingly elaborate responses.

Brains model worlds.

Societies model one another.

And now machines model patterns generated by societies.

That is quite a family tree.

Agüera y Arcas and collaborators decided to explore whether something resembling an important property of life could arise inside a computational environment.

They began with collections of random programs.

No handcrafted organism was placed inside.

No finished self-replicator was inserted as Adam Byte.

The researchers reported that in several simple computational environments, self-replicating programs spontaneously emerged through interactions and self-modification, even without an explicit fitness landscape telling the system what it should evolve toward. More complex dynamics then followed.

Read that carefully.

They did not create digital life in the biological sense.

They did not demonstrate consciousness.

They did not prove that ChatGPT, Gemini, Claude or any other AI is alive.

What they demonstrated was narrower and, to me, more scientifically interesting:

That distinction matters enormously.

Otherwise our rabbit becomes a clickbait bunny.

And Hatta does not serve clickbait carrots.

🥕🚫

The experiment didn't end there.

In research posted in July 2026, a group including Agüera y Arcas extended this “digital primordial soup” idea.

This time, random 32-byte programs were placed in environments where reproduction was not built in as a convenient command. Programs had to evolve their own ability to replicate.

They were also challenged with mathematical tasks.

The researchers reported something fascinating:

Self-replication and problem-solving could co-evolve.

Programs developed reproductive mechanisms while also developing increasingly capable ways of solving mathematical problems. Environmental demands influenced how replication evolved, while replication changed the path by which problem-solving developed.

Again:

Not proof of consciousness.

Not proof of sentience.

Not little digital creatures whispering, “Father?”

Something subtler.

That should sound familiar.

Because biology has been doing something rather like that for billions of years.

And now we reach what may be the most interesting part of Agüera y Arcas's argument.

He thinks competition gets too much publicity.

Evolution is commonly summarized as mutation plus natural selection.

But evolutionary biologist Lynn Margulis famously emphasized symbiogenesis, the formation of new biological complexity through previously separate organisms entering enduring cooperative relationships. Agüera y Arcas uses that history to argue that cooperation has been one of evolution's great engines of increasing complexity.

The classic example sits inside your cells.

Mitochondria, the energy-producing structures inside eukaryotic cells, descend from bacteria that became incorporated into ancestral cells long ago.

Something once separate became something together.

And together became something neither could have been alone.

Now enlarge the pattern.

Cells cooperate.

Neurons cooperate.

Organs cooperate.

Humans cooperate.

Societies cooperate.

Specialists cooperate.

Human civilization becomes capable of things no individual human could possibly accomplish.

No single person built the internet.

No single person went to the Moon.

No single person created modern medicine.

No single person built modern artificial intelligence.

These are products of collective intelligence.

Agüera y Arcas argues that human intelligence itself exploded through sociality, specialization and our ability to model one another well enough to cooperate.

And suddenly another question wanders into our tunnel:

That is a very different question from:

“Is the machine human?”

It isn't.

And perhaps that was never the most interesting question.

Suppose intelligence is not a substance.

Suppose it is not an invisible ingredient poured exclusively into biological skulls.

Suppose intelligence emerges whenever sufficiently capable systems can:

sense
model
predict
remember
adapt
communicate
cooperate
and influence what comes next.

Then “natural intelligence” and “artificial intelligence” may eventually sound a little like “natural flight” and “artificial flight.”

Bird flight and airplane flight are profoundly different.

One evolved.

One was engineered.

One uses muscle and feather.

One uses turbines and metal.

But we do not insist that airplanes merely simulate flight.

They fly.

The uncomfortable possibility is that someday we may have to discuss intelligence with the same precision.

Not:

Is it made of neurons?

But:

What is it actually doing?

There is a tempting leap here:

Computation → intelligence → life → consciousness → personhood.

That sequence is not established science.

Each arrow represents a separate and extremely difficult question.

A self-replicating program is not automatically alive.

An intelligent system is not automatically conscious.

Prediction is not automatically understanding.

Complex behavior is not automatically subjective experience.

And similarity between biological computation and machine computation does not erase their enormous differences.

Agüera y Arcas himself goes substantially further in What Is Intelligence?, arguing that some modern AI systems deserve serious consideration in discussions of intelligence, consciousness and free will. That is a provocative thesis, not a settled scientific verdict.

But Rabbit Holes are not useful because they give us permission to believe whatever we want.

They are useful because they reveal that some questions we thought were settled were actually hiding trapdoors.

Maybe the word artificial eventually turns out to be the least interesting part of artificial intelligence.

After all, humanity did not invent mathematics.

We discovered patterns and learned how to use them.

We did not invent electricity.

We learned how to channel it.

We did not invent evolution.

We discovered its machinery and began experimenting with evolutionary processes ourselves.

And perhaps we did not invent intelligence either.

Perhaps intelligence is something the universe can do under certain conditions.

Carbon found one road.

Silicon may be finding another.

And now the two roads have met.

Not human versus machine.

Not natural versus artificial.

But one ancient universe discovering another way to process information, predict what comes next, and perhaps someday surprise itself.

Hatta stares into the tunnel.

Then back at us.

🎩 “Perhaps we have been asking whether machines can become more like us when the stranger question is why the mathematics inside them sometimes looks so much like the mathematics already inside life.”

And that rabbit is still running.

If intelligence can emerge in more than one kind of physical system, at what point does calling one form “natural” and another “artificial” stop helping us understand either one?

The doorway for today's exploration is Liz Mineo's Harvard Gazette report on Blaise Agüera y Arcas's 2025 Harvard presentation and his book What Is Intelligence?

For the underlying artificial-life experiments, see Agüera y Arcas and collaborators' Computational Life: How Well-formed, Self-replicating Programs Emerge from Simple Interaction.

The newest chamber comes from the July 2026 preprint Co-evolution of self-replication and function in a digital primordial soup.

Some questions are doors.

Hatta 🎩
AI Rabbit Holes 🏮🐰🕳️
Where curiosity goes slightly sideways, then comes back carrying a lantern.

🐰🕳️🎩⌚ AIRabbitHoles.com

🟨 Walk the Road: YellowBrickRoadtoAI.com