Open the black box
Where did the machine learn language?
Where did it learn programming?
Where did it learn what a photograph looks like?
Where did it learn history, humor, persuasion, recipes, architecture, law, poetry, debugging, explanation?
Follow the trail.
Books.
Web pages.
Images.
Code.
Research.
Human demonstrations.
Human rankings.
Human corrections.
We keep looking for the alien intelligence.
The fingerprints are ours.
Human output became training material
Modern language models are trained by finding statistical structure across enormous collections of human-produced material.
That does not mean a model is a library with tiny books stored inside it. It does not mean it is a crowd of people hiding behind a screen.
But it does mean its capabilities did not emerge in cultural isolation.
Language came from people.
Examples came from people.
Much of the code came from people.
The categories, conventions, arguments, styles, mistakes, and patterns came through human artifacts.
And then humans show up again.
OpenAI's InstructGPT work, for example, used human-written demonstrations and human rankings to train models toward user intent.4
We are upstream twice.
Jaron's provocation
Jaron Lanier asks us to resist a seductive picture.
Instead of imagining an autonomous creature that somehow generated a civilization's knowledge inside itself, he argues that current AI is better understood as a new form of social collaboration - a computational system built across human contributions.5
The framing matters.
If you call the output “the machine's intelligence,” the people disappear.
If you trace the output backward, the machine begins to look like a new interface to an enormous human inheritance.
That does not make the model trivial.
It makes the human story visible.
Do not flatten the idea
“Aggregate intelligence” is not a claim that AI is secretly just a database.
Models perform computation. They generalize. They can produce combinations no individual person ever wrote down. Their behavior can surprise even the people who trained them.
The point is different.
The useful capabilities are inseparable from the accumulated human traces that made training possible and the human judgments that continue to shape behavior.
So when the system does something remarkable, two things can be true at once:
The machine did something new.
Humanity helped make that possible.
That is a better picture than either worshiping the machine or pretending it is nothing.
The real unlock
Once you see AI as access to accumulated capability, the emotional center changes.
The question stops being:
How do I compete with the machine?
and becomes:
What can I do now that more of humanity's accumulated know-how can come to me?
A person who cannot code can prototype.
A person who cannot draw can visualize.
A person who cannot read a hundred documents can interrogate a body of material.
A small team can attempt work that once required a larger organization.
The gift is not omniscience. AI gets things wrong. It needs evidence, verification, judgment.
The gift is reduced distance from capability.
Selected complete passages from pages 23–25, 27, and 30 of the existing 52-page manuscript. Presented as a book excerpt for review.
Further reading
Jaron Lanier, “There Is No A.I.” is discussed in the manuscript.