When Curiosity Becomes Cheap
AI may matter less because it can answer our questions than because it has radically reduced the cost of pursuing them
Bill Bonner wrote something today that got me thinking.
His argument, stripped down, is that artificial intelligence has no privileged access to truth. Large language models are built from human words, human ideas, human judgments and human mistakes. AI may be clever, but it is still standing on a mountain of material produced by us.
Fair enough.
But I think that misses something important.
Perhaps profoundly important.
AI is very good at connecting dots.
And human beings, for all our intelligence, have become exceptionally good at putting those dots into separate boxes.
Medicine is divided into specialties. Science into disciplines. Universities into departments. Research into journals. Economics, history, engineering, biology, physics, politics — each develops its own vocabulary, experts, literature and assumptions.
The knowledge may exist.
The connection between two pieces of knowledge may exist.
But nobody necessarily sees it.
Because nobody is looking at both.
AI doesn’t particularly care which department owns the information.
And that may change far more than we realize.
The old cost of curiosity
I grew up with the World Book Encyclopedia.
If something caught my attention, that was where I started.
World Book could take me surprisingly far. But eventually I would reach the edge of the entry.
Then came the library.
The library could take me farther still, assuming I knew what to look for and could find the right book.
Eventually, though, serious curiosity often ended at a textbook.
And textbooks were formidable things.
Six hundred pages. Eight hundred pages. Dense terminology. Long chapters establishing concepts you might need eventually but didn’t particularly care about now.
Somewhere inside that book might be the answer to the question that had originally fascinated you.
But there was a substantial admission price:
Time. Background knowledge. Patience. Access. The ability to understand the terminology. The ability to figure out which parts mattered.
And quite often, the question simply died.
Not because you stopped being curious.
Because pursuing it cost too much.
That is what AI has changed for me.
I can begin with the interesting part.
Why?
Then:
How?
Then:
What would happen if...?
Then:
Wait. If that’s true, wouldn’t this other thing happen too?
And suddenly we are six levels deeper than the original question.
Sometimes much deeper.
I have had more than a few conversations that started with an innocent question and ended, much later, with me staring at the screen thinking:
Holy shit.
The remarkable thing is not merely that AI supplied information.
It removed friction.
The textbook hasn’t disappeared
There is an obvious danger here.
Sometimes the boring parts matter.
Sometimes the chapter we would rather skip contains the definition, mathematical constraint, methodological detail or inconvenient evidence that destroys our beautiful idea.
AI cannot make rigor optional.
If anything, easier hypothesis generation makes rigor more important.
Eventually you may still have to read the paper. Check the data. Understand the experimental design. Look at the primary source. Do the math.
But the route to that material has changed.
Instead of:
curriculum → facts → understanding → interesting question
we can increasingly work like this:
interesting question → explanation → missing concepts → targeted reading → better question
That is a very different way of learning.
The textbook becomes less of an obstacle course and more of a tool.
Instead of starting at page one, we can start near the intellectual frontier of the particular question that interests us, then work backward into the foundations we actually need.
And, crucially, we can interrupt.
A book cannot respond when paragraph three makes you say:
“Hold on. Why?”
AI can.
Then I thought about a cardiologist
A few days ago I was with an older relative at her cardiologist.
She had been taking a medication that was leaving her very tired in the morning. He switched her to something else, and we discussed how she was doing.
At some point I asked whether he might be willing to read a paper I have been working on about diuretics, dehydration and the difficulty of determining volume status in older patients.
He said he would love to.
But he barely has time for his family.
Never mind finding time for a paper on a complicated subject he hadn’t expected to encounter that day.
And looking at him, I believed him.
He is younger than I am.
He looked exhausted.
Think for a moment about the information burden carried by a physician like that.
Patients. Medications. Blood tests. Echocardiograms. Guidelines. Drug interactions. Referrals. Follow-ups. Chart notes. New research. Old research. Administrative work.
And behind all of it, the knowledge that the decisions being made aren’t academic.
There are human lives attached to them.
Then somebody like me wanders into the office and says:
“Here. I think there may be something interesting in this paper.”
The problem isn’t that he doesn’t care.
The problem is bandwidth.
Now imagine giving that physician an AI system capable of reading the paper first.
Not deciding whether it is correct.
Not practicing medicine for him.
Not stamping the hypothesis VERIFIED.
Simply saying:
Here are the five claims most relevant to cardiology.
Here is the evidence offered for each.
Here are the assumptions.
Here are the weak links.
Here is where the argument agrees with or challenges existing literature.
Here are the strongest counterarguments.
Here are the three sections worth reading yourself.
Here are the primary papers behind them.
Now he can ask:
“Why does the author think this?”
Then:
“Show me the evidence involving patients over 80.”
Then:
“What about those with chronic kidney disease?”
Then:
“Show me the strongest evidence against the hypothesis.”
Then:
“What would we have to measure to determine whether this is actually happening?”
He hasn’t verified the hypothesis.
He hasn’t reproduced the investigation that created it.
He doesn’t need to.
AI has compressed the route sufficiently for him to decide whether the destination is worth examining.
Instead of rebuilding the argument from raw material, he gets to begin with a better-formed hypothesis and interrogate it.
The hypothesis itself must still survive contact with reality.
That may require prospective observation, better measurement, an experiment, replication or years of accumulated evidence.
AI doesn’t remove that.
What it removes is much of the excavation required merely to get there.
That isn’t replacing his judgment.
It may be restoring his ability to use it.
Compression may matter as much as intelligence
This is where I think much of the conversation about AI goes wrong.
We keep asking whether AI is intelligent.
Whether it has judgment.
Whether it knows truth.
Whether it can replace professionals.
Those are important questions.
But perhaps another capability matters just as much:
compression.
AI can reduce enormous bodies of information into something a human being can realistically examine while preserving the ability to drill back down into the source material.
That changes the economics of thought.
The physician doesn’t need another machine confidently telling him what to do.
He needs help getting 40 relevant pages down to four paragraphs — with the important qualifications intact — so that his scarce attention can be spent where it has the greatest value.
There is an instructive historical precedent.
The printing press didn't make Aristotle's ideas any truer.
It didn’t remove error, superstition, bad ideas or outright nonsense from human thought.
What it changed was the cost of reproducing ideas.
And once copying became cheaper, other things became easier too. Texts could be compared. Arguments could be checked against other arguments. Ideas could travel farther. More people could enter the conversation. Knowledge that had once been geographically or institutionally separated could increasingly be placed side by side.
The printing press made ideas cheaper to reproduce.
AI makes them cheaper to traverse.
It changes the cost of navigating, interrogating and recombining knowledge in something approaching real time.
The printing press didn’t improve the stock of human knowledge.
It transformed the flow of it.
AI may be doing the same thing at another layer: making knowledge dramatically easier to navigate, interrogate and recombine.
Bill looks at the accumulated corpus on which AI rests and quite reasonably sees the trash heap mixed in with the treasure.
But perhaps he is looking at the stock and missing the flow.
He can be entirely right about the trash heap and still miss the revolution.
The important change may not be what knowledge AI contains.
It may be what becomes possible when the cost of moving through, comparing, questioning and recombining that knowledge collapses.
A historian can cross-reference economics.
An engineer can wander into physiology.
A retired businessman can start asking questions about renal function, blood rheology, endothelial shear stress, monetary history, nuclear physics or whatever else happens to catch his attention that morning.
The traditional barriers remain.
Expertise still matters. Evidence matters. Experience matters.
And some knowledge is not written down at all.
The tacit judgment that comes from years of practice — knowing what looks wrong, what matters, what usually fails, what doesn’t quite fit — does not compress nearly as easily as text.
AI can help a non-specialist reach the edge of a question much faster.
It cannot instantly confer twenty years of experience standing at that edge.
The walls have become much lower.
They have not disappeared.
The economics of curiosity
Somewhere in thinking about all of this, one sentence stopped me:
What happens to human curiosity when the cost of pursuing a question collapses?
That may be the thing.
For most of human history, curiosity was expensive.
Every question carried a tax.
You needed access to information, then you needed to find it, understand its language, determine what mattered, locate the next source and perhaps find an expert willing to explain what you still didn’t understand.
Most questions never survived the process.
They evaporated somewhere between:
“That’s odd...”
and:
“I wonder why?”
AI has dramatically reduced that tax.
Not eliminated it.
Reduced it.
And when the cost of pursuing curiosity falls, more questions survive.
More questions surviving means more connections.
More connections mean more hypotheses.
More hypotheses mean more opportunities to discover that something everyone has been looking at separately may belong together.
But there is an important asymmetry.
AI has made verification cheaper too.
A literature search that once required days can sometimes be done in hours. Claims can be traced to sources. Conflicting papers can be found more quickly. Several explanations can be compared. A model can even be asked to attack its own conclusion.
That is real progress.
But hypothesis generation has become cheaper even faster.
We can now ask:
“Could A explain B?”
a hundred times before lunch.
Many of those ideas can also be rejected quickly.
But establishing that one of the survivors actually reflects reality may still require careful measurement, primary evidence, specialist expertise, experimentation, replication or time.
So the important change isn’t that verification remains expensive while curiosity becomes free.
It’s that the ratio has shifted.
The cost of generating plausible explanations is collapsing faster than the cost of proving them.
That creates abundance.
And a new bottleneck.
AI can produce an elegant bridge between two islands that aren’t actually connected.
It can also faithfully summarize five papers that all share the same mistaken assumption.
Asking five AIs the same question may help reveal whether one model misrepresented a source.
It does not magically turn five models trained on overlapping human knowledge into five independent experiments.
Consensus is not the same thing as truth.
So discernment matters enormously.
Is this connection interesting?
Is it merely plausible-sounding?
What evidence would change my mind?
What observation could prove this wrong?
Are my sources actually independent?
When do I stop asking the machine and measure something in the real world?
Those may become some of the most valuable questions of the AI age.
For generations, some of the friction in learning acted as an accidental brake. You couldn’t race very far ahead of your understanding because sooner or later the textbook, the mathematics or the laboratory stopped you.
AI removes much of that natural braking mechanism.
That is wonderful for curiosity.
But it may mean we need to develop deliberate brakes where friction once supplied them accidentally.
Not barriers to asking questions.
Barriers to becoming prematurely certain of the answers.
AI doesn’t eliminate human judgment.
It may increase the demand for it.
The machine can generate possibilities almost without limit.
Someone still has to decide which possibilities deserve to survive.
The human part
There is another reason I don’t think this ends with AI doing our thinking for us.
AI needs something from us.
Lived experience.
The odd observation.
The thing that bothered us.
The memory of an exhausted cardiologist.
The question that didn’t fit.
The instinct that two things somehow belong beside each other.
I can feed an AI those experiences.
And something interesting happens.
It can hold the pieces together and look for structure.
I supply the dots.
It helps draw lines between them.
Then I look at the picture and decide whether the lines make sense.
That leads to another question, which becomes another dot, then another connection.
The process begins to look less like:
AI thinks → human receives
and more like:
human notices → AI connects → human judges → AI investigates → human notices something new
Round and round.
That is a very different relationship.
Stranded Human Curiosity
And this is where the potential becomes enormous.
There must be millions of people carrying questions they never pursued.
Not because they lacked intelligence.
Not because the questions were foolish.
Because they weren’t doctors, physicists, economists, engineers or academics.
Because they didn’t know the terminology.
Because they didn’t know where to start.
Because nobody had time to explain it to them.
Because opening an 800-page textbook seemed like a ridiculous commitment merely to satisfy a suspicion that something didn’t quite add up.
We often talk about stranded capital, stranded assets and stranded talent.
I wonder how much stranded human curiosity has accumulated through history.
Questions were noticed and abandoned, connections glimpsed but never pursued. Nurses, mechanics, farmers, patients, technicians, businesspeople and grandparents understood one small corner of the world but lacked the vocabulary, access, credentials or time to follow the question into another.
We have no way to measure how much potentially useful thought died there.
AI may not democratize expertise.
Expertise takes work.
Experience still takes time.
Reality does not waive its standards merely because we can ask better questions faster.
But AI may democratize something that comes before expertise.
The ability to begin an investigation.
That matters.
Because now someone can start with four words:
This doesn’t make sense.
And keep going.
They can ask the next question.
And the next.
And the next.
They can discover what terminology they were missing, find the field that owns the problem, encounter the counterargument and learn enough to realize their original idea was wrong.
Or, occasionally, discover that the odd little thing they noticed really does deserve another look.
Perhaps the coming awakening isn’t that millions of people discover artificial intelligence.
Perhaps it is that millions of people discover their own ability to investigate things.
People who spent their lives thinking:
“I’m interested in this, but I’m not a doctor.”
“I’m not a scientist.”
“I wouldn’t know where to begin.”
may suddenly realize that beginning is no longer the hard part it once was.
And somewhere far downstream from the innocent question they originally asked, they may find themselves staring at the screen thinking:
Wait. I can actually think about this.
Bill may be right that AI possesses no divine source of truth.
It doesn’t.
Neither do we.
But that may not be the most important thing happening.
The revolution may not be that machines have acquired judgment.
It may be that ordinary human curiosity has suddenly acquired extraordinary leverage.
The stock of human knowledge remains messy, contradictory, brilliant, foolish, inspired and wrong.
Just as it always was.
What has changed is our ability to move through it.
To interrogate it.
To connect it.
To challenge it.
And to follow a question far enough that another question appears.
David Deutsch argues in The Beginning of Infinity that the growth of knowledge has no obvious endpoint: explanations lead to new problems, and answers make previously impossible questions possible.
AI doesn’t change that process.
What it may change is how many more people can participate in it — and how quickly they can reach the next question.
We have no idea yet what happens when millions of people discover that they can afford to take their own curiosity seriously.
But I suspect we’re barely on page one.
Regards,
Luke Kandia
Luke Kandia writes Chasing Clarity — one man’s search for a better understanding of the world, both small and large, around him.
An interesting view of AI. This I would not have thought of by Myself. Thanks for enlarging this tool in My tool box.