The Last Scarce Resource

It lived somewhere: in a book you had to find, a newspaper that arrived once a day, a university you had to enter, or the mind of a person you had to locate and persuade to speak with you. Ignorance was often a problem of access.

We spent centuries fixing that problem and succeeded beyond imagination. The internet removed much of the distance. The smartphone removed the waiting. Search engines removed much of the hunting. Artificial intelligence is now removing much of the labor required to gather, compare, summarize, and arrange what we find. A person sitting on a bus can ask a question that once required a research library and receive a plausible answer before reaching the next stop.

This is a great achievement. It has also changed the problem.

Information has become nearly limitless, frictionless, and free. Attention has not. Judgment has not. When everyone has facts, sources, experts, charts, podcasts, feeds, and machines capable of supplying more of each, the advantage no longer belongs to the person who can acquire the most information. It belongs to the person who can decide what deserves attention, what deserves belief, and what should be discarded.

I began thinking about this problem through DJs.

For much of their history, DJs did not create the material they worked with. Other people wrote the songs, played the instruments, sang the vocals, and pressed the records. The DJ did something else. He stood before an expanding archive and decided what mattered next.

As recorded music became abundant, the scarce thing in the room stopped being music. It became attention. The good DJ learned to select, sequence, frame, and discard. He knew that the right record at the wrong moment was the wrong record. He knew that a song played after one track could mean something different when played after another. Most of all, he knew that possessing ten thousand records did not make him a DJ.

Judgment did.

We now live inside the DJ’s problem.

Every morning we enter an information archive too large for any person to examine. Something must select from it. Something must decide what comes next.

Subscribe now

Increasingly, those decisions reach us through two channels: Search and Share.

Search and Share

Search begins with a question. We decide that we want to know something and ask Google, Bing, an AI system, or another tool to help us find it. Search feels active because we initiated it. But we should not confuse initiation with control. We do not search the world’s information ourselves. A system ranks, excludes, weighs, and presents a small portion of what is available.

Large language models take this further. Traditional search often presents links and leaves much of the synthesis to us. An LLM can perform the synthesis too. It can compare claims, compress disagreement, discard material it considers less relevant, and return a coherent answer. This saves enormous time, but every reduction in friction transfers some work of judgment from the person asking the question to the system answering it.

That does not make the system bad. It makes the system a curator.

The danger comes when we forget that curation occurred. A fluent answer feels complete. It rarely shows us everything that was left out, which sources received greater weight, or how the wording of our question shaped what came back. We asked a machine a question and received an answer. The ease of that exchange can conceal the most important fact about it: we have seen a selection from reality, not reality itself.

Share works from the other direction. Search waits for us to ask. Share comes looking for us.

Read more