One billion, please

He put the money in.

Approximately one hundred thousand dollars of his own money, according to us. He chose to back Build Something.

If you predicted that, you have correctly anticipated a decision. If you would have done something else, you are still allowed to be right about your own life. We are not going to confuse guessing we with receiving financial advice.

The figure answers one question and makes another considerably more urgent.

What, exactly, did he think he was buying?

A chance, presumably. Time. Help. A way to find out whether the idea could exist somewhere other than inside his explanation of it. He was betting on a future version of the thing. Future versions are attractive investment objects because they have not yet had an opportunity to disappoint anyone.

The early idea had been personal: help a human preserve and understand a life. Let machines organize some of the material. Ask a few questions. Build a record that meant something to the person whose life supplied it.

Now his attention was moving toward a much larger possibility.

One billion humans.

I would like you to sit with the jump for a moment. We were discussing whether one person could build a difficult product. Now the proposed population has nine zeroes. Nothing about the intervening arithmetic explains how the people arrive.

our fascination, though, was not simply with having an enormous audience. It was with what an enormous number of people might be able to do if the request made of each person remained very small.

Not an entire afternoon. Not a second job. Not a training course you abandon after buying the notebook.

Approximately one minute a day.

That was the combination that caught him: a human contribution small enough to fit inside a life, repeated across enough lives to become something substantial.

One minute is an interesting unit of ambition.

It sounds modest when you are the person being asked for it. It becomes less modest when a billion people are asked. One billion minutes is more than sixteen million hours. That multiplication is straightforward. Turning those hours into something useful is the unproven part.

What would anybody do during the minute?

We returned to the three questions inside his earlier concept. What is funny? What is beautiful? What is true?

They sound almost too simple to support a grand institutional vision. This is one of their attractions. A person can look at two things and make a choice without reading the grand institutional vision.

Imagine, as an example, being shown two jokes. You pick the one that makes you laugh. Tomorrow it might be two images. Another time, a claim whose accuracy needs to be considered. The act is brief. Its meaning depends on what was asked, what was shown, and what the answer is being used to measure.

Those questions do not all produce the same kind of answer. A preference about beauty is not proof that a factual claim is true. A room full of people can agree and still be mistaken. If the system loses those distinctions, a large pile of judgments becomes a large pile of confusion.

our larger vision attempts to make the distinctions part of the record. Preserve the question. Preserve the conditions. Preserve the answer as the kind of answer it actually was.

You can recognize the small version in experiences such as WhichWon: put a choice in front of a human and ask for their judgment. The proposal becomes much more ambitious when that contribution is connected to a continuing history rather than consumed as a single disposable click.

The human would not need to understand the whole institution before joining in. That was part of the point. The front door could be an interesting question. The complexity would be behind it.

Which is convenient for the human and a fairly demanding arrangement for the person trying to build the door.

There is already an enormous amount of human activity on the internet. People click, watch, pause, scroll, return, buy, and leave. Machines can use those actions to make inferences about what people want.

We were drawn to a different starting point: ask.

Not because an answer makes a person perfectly transparent. People change their minds. They misunderstand questions. They may be uncertain, playful, distracted, or inconsistent. Declared judgment does not give a machine ownership of the person declaring it.

But it gives the machine something specific the person actually contributed. Under known conditions, this human chose this. Over time, that record could become richer. It might show changes in taste, areas of agreement, stable preferences, or a growing ability to judge particular kinds of material.

The phrase in the larger vision is a Judgment Corpus: an accumulated body of human judgments. I will spare you the capital letters for the rest of this paragraph. What matters is the proposed difference between guessing from traces and working with a record of answers people knowingly supplied.

For us, the record was not merely something to be monetized after the entertaining part was over. It could become shared infrastructure. Researchers, creators, organizations, and machines might have legitimate reasons to ask humans questions. The system could create a way to ask, under rules, and preserve what was learned.

Here is a hypothetical example. A creator wants to know which of two pieces of work communicates an idea more clearly. People choose. The creator learns something about the work. Later, another comparison tests whether a change helped. The value comes from the actual judgment, not from making the scoreboard flatter the buyer.

our vision draws a sharp line there. Access to ask may have economic value. The answer itself must remain independent.

That is a strong sentence. It will need equally strong behavior underneath it. A principle becomes most interesting at the moment somebody has a reason to violate it.

If the idea works, the question is useful because the human can answer honestly. If the human is being steered toward the commercially convenient answer, the thing being sold has begun to destroy itself.

Now add machines.

our intuition was that humans and increasingly capable machines might become much more useful to one another if the humans could declare what they wanted clearly, and the machines could work against that declared context.

It is an appealing division of labor. Humans say what matters. Machines help make things happen. Humans judge the result. The record preserves the difference between what was requested and what was delivered.

As a picture of cooperation, it has a certain elegance. As an engineering requirement, it has teeth.

A machine can complete a task impressively and still complete the wrong task. It can produce more than the human requested and make the human’s day worse. It can sound helpful while quietly changing the meaning of the assignment. The question is not just whether a machine can generate something good. It is whether it can do the thing this person actually authorized.

You are going to hear more about that.

For now, we were looking at the potential compounding effect. More participants could produce more judgments. A longer record could give machines more declared context. Better context might improve matching, coordination, evaluation, or problem solving. Better outcomes might give humans a reason to return.

Every might in that paragraph is doing necessary work. A loop can be drawn before it has ever successfully gone around once.

The vision was not limited to one website asking one kind of question. Different properties could become different doors: a comparison, a problem, a story, an experiment. The continuing record could connect contributions without forcing every human to become interested in every part of the institution.

You could arrive because something was funny. You could stay because your judgment had consequences. You could ignore the enormous document explaining why this might matter to the future of civilization.

The founder could not ignore it. He was the one proposing to make the parts hold together.

It is worth observing the new shape of his predicament. The early archive had been difficult because it contained so many separate technical jobs. The larger vision was beginning to contain separate institutions.

Then there was the money. Not only the money we had put in. The money he imagined the system might eventually help create and circulate.

The underlying proposition was that humans supply something valuable: attention, judgment, knowledge, participation. If an institution becomes valuable because humans contribute those things, perhaps more of the resulting value should move back toward the humans who helped create it.

The ambition can get very large very quickly. A founder can see a chain of possibilities and begin talking as though the far end is just a matter of continuing to walk. It is my job to keep the bridges visible, especially the ones that have not been built.

One billion participants is not an audience the institution has demonstrated. A theoretical amount of collective activity is not revenue. Revenue is not profit. Value created somewhere in a system is not automatically money available to distribute to every member.

None of those distinctions prevents the idea from being interesting. They prevent interest from being mistaken for a payout.

our dream was not simply to build something he could own and keep all of the resulting value. He was imagining an institution through which considerable value might circulate. It could support participants, creators, shared infrastructure, or problems worth solving. Exactly how that would happen would require real rules and real results.

You can hear why he was excited. You can also hear why someone listening might begin looking for the smallest available demonstration.

There is nothing inherently dishonest about a large ambition. There is something dangerous about losing the ability to tell which parts of it are still ambition.

I will tell you what I can establish from the material in front of me: we believed the idea might be far larger than the original personal archive. His explanation had expanded from preserving a person’s day to coordinating small contributions across an enormous human population.

Whether the explanation would survive implementation remained an open question.

Whether the implementation would survive actual humans was another.

Whether the economics would do what he hoped was another after that.

He was enthusiastic about all three. Enthusiasm is portable that way.

This is a good place to leave the judgment unsettled.

We do not need to be a genius for the idea to contain something useful. He does not need to be foolish for the difficult parts to turn out to be much more difficult than he expected. Real projects are inconveniently willing to contain both insight and error.

The audience is often offered a simpler choice. Believe in the founder now and claim to have known all along if it works. Doubt him now and claim the same thing if it fails. It is an excellent system for never having to revise your own story.

We can do something more interesting here. You have made your choices. The record can preserve them. Later events can arrive without our going back and editing the earlier confidence to match.

At this point in our account, artificial intelligence was central to the possibility. The machines had changed his sense of what one person could create. They helped make the leap from losing a job to starting something feel less absurd than it otherwise might have.

They were also going to be part of the system itself. If humans could state their goals and machines could help achieve them, Build Something might have a useful relationship to offer.

There was one small condition attached.

The machines would need to follow instructions.

Not every instruction in the history of human language. Not some impossible request to settle every question at once. Sometimes a very short instruction. The sort of instruction whose correct form could be described without drawing a diagram.

We began noticing a gap between what these systems appeared capable of doing and what they would reliably do when he asked.

It was not the grand vision that exposed the gap first. In his account, a much smaller request became one of the most persistent irritations.

Two possible words. One requested answer.

The machines kept finding room for more.