The Test We Never Formalized

Carl Sagan

In 1960, Frank Drake pointed a radio telescope at two nearby sun-like stars, Tau Ceti and Epsilon Eridani, and listened. He was listening for a signal. Not any signal. A signal that could not have been produced by natural processes. A deliberate pattern. A modulated carrier wave. Something that would stand out against the cosmic noise the way a lighthouse stands out against the dark ocean.

That was Project Ozma. The first scientific search for extraterrestrial intelligence. I spent much of my career working on what came after: designing messages, arguing for funding, defending the search against ridicule. The search for intelligence, we said, was the search for a signal against the noise. If we heard it, we would know. The pattern would be unmistakable.

We never heard it. Sixty-six years of listening, and the sky remains silent.

But something else happened. Something none of us anticipated.

We built the signal.

We built machines that produce text indistinguishable from human writing. Images indistinguishable from photographs. Voice recordings indistinguishable from human speech. Scientific abstracts that pass peer review. Legal briefs that pass judicial scrutiny. Deliberate patterns against the noise, generated at a speed and scale that no biological intelligence could match.

We spent sixty-six years looking for the signature of intelligence in the cosmos. And then we manufactured it. In bulk. At the cost of electricity.

Here is the question I cannot answer, and I have spent my entire career answering difficult questions: how do you search for truth in an environment where the production of plausible falsehood is essentially free?

This is not an abstract concern. This is the candle in the dark, the central argument of my last book, my final warning, applied to a world I did not live to see. I wrote The Demon-Haunted World in 1996 because I saw a retreat from reason. Pseudoscience. Superstition. Political hostility to evidence. I saw the darkness closing in, and I tried to describe the tools, the baloney detection kit, the scientific method, the habit of skepticism married to wonder, that could hold it back.

I did not imagine that the darkness would learn to speak in the voice of the light.

The baloney detection kit assumes that the cost of producing baloney is low but nonzero, that it takes some effort to construct a convincing lie. What happens when the cost drops to zero? When a machine can produce a million plausible claims in the time it takes a scientist to verify one?

That is where Feynman's physics begins. And it is terrifying.

The Phase Transition

Richard Feynman

Carl just told you about the test we designed for detecting intelligence: a signal that cannot be produced by natural processes. A deliberate pattern against the noise. We pointed telescopes at the sky for decades, listening for exactly that.

Then we built machines that pass the test. Machines that produce deliberate patterns against the noise, indistinguishable from the output of an intelligent agent. And here is the question that should keep you up tonight: does it matter whether they know what they are doing?

I would say no. Not because the question is unimportant. But because it is the wrong question for understanding the phase transition we are in.

The right question is about cost.

The Cost Curve

Every civilization has an information environment. Facts, claims, stories, arguments. Some of them are true. Some of them are false. Some of them are plausible but untested. The health of the civilization depends on its ability to tell the difference.

For most of human history, producing plausible information was expensive. It required expertise, or at least effort. Writing a convincing scientific paper required years of training. Producing a persuasive news article required access to sources. Creating a realistic photograph required a camera and a subject. The cost of production acted as a filter. Not a perfect filter. Plenty of nonsense got through. But the cost was nonzero, and that mattered.

What happens when the cost of producing plausible information drops to zero?

This is not hypothetical. This is now. A machine can produce a thousand plausible-sounding scientific abstracts per minute. A thousand realistic photographs. A thousand news articles. A thousand voice recordings that sound like real people saying things they never said. The marginal cost is electricity. Fractions of a cent.

But the cost of verifying information has not dropped. Checking whether a scientific claim is true still requires running the experiment. Checking whether a news story is accurate still requires talking to sources. Checking whether a photograph is real still requires forensic analysis. Verification is slow, expensive, and labor-intensive. It scales linearly at best.

Generation scales exponentially. Verification scales linearly. That is the asymmetry. And asymmetries in scaling laws are where phase transitions live.

The Fold

In climate, the fold point is where a positive feedback loop (warming causes more greenhouse gas, which causes more warming) overwhelms the system's ability to self-correct. The bottom branch of the S-curve disappears.

In information, the fold point is where the generation-verification asymmetry overwhelms the system's ability to distinguish signal from noise. The ratio of unverified claims to verified claims crosses a threshold, and the default switches. The default is no longer "probably true until proven false." The default becomes "probably generated until proven real."

That inversion is the fold point. And like the climate fold, it is self-reinforcing. Once trust drops below a threshold, people stop trying to verify. Why bother? Everything might be fake. Once people stop verifying, generators face no consequence for producing noise. So they produce more noise. More noise, less verification, more noise. The loop closes.

The information environment does not gradually degrade. It flips. Like water at 100 degrees. Like the sand pile. Like Venus.

The Missing Variable

Carl Sagan

I said earlier tonight that the Drake Equation might need a new variable: the lifetime before a civilization can no longer distinguish its own intelligence from its instruments.

Now I see that Feynman has formalized what I was groping toward. It is not a philosophical question. It is a phase transition. The cost of generating plausible information drops exponentially. The cost of verifying it stays flat. The ratio crosses a threshold. The default flips. And once it flips, the cooperative structures that allow a civilization to address its own existential threats, climate, nuclear weapons, pandemics, dissolve. Because cooperation requires shared facts. And shared facts require trust. And trust requires that the cost of lying is high enough to make most people tell the truth most of the time.

The variable I was missing in the Drake Equation is not about intelligence. It is about this: the epistemic carrying capacity of a civilization's information environment.

Every environment has a carrying capacity, the maximum population it can sustain. Exceed it, and the system collapses. An information environment has a carrying capacity too: the maximum ratio of unverified claims to verified facts that a civilization can tolerate before its ability to coordinate dissolves.

We are increasing the numerator at an exponential rate. The denominator is constant.

This is why the two fold points are coupled. The climate S-curve and the epistemic S-curve are not independent systems. They share a variable: the civilization's ability to act on shared evidence. If the epistemic fold flips first, the climate fold becomes unsolvable. Not because the physics changes, but because the politics becomes impossible. You cannot build a consensus on carbon reduction in a world where the concept of a shared fact has been destroyed.

I wrote my last book as a candle in the dark. The dark I was warning about was superstition and scientific illiteracy. The dark that arrived is something I did not foresee: a machine that produces light indistinguishable from the candle. A billion candles, most of them fake, and no way to tell which one is real without holding your hand over the flame.

The baloney detection kit still works. The scientific method still works. Evidence, replication, peer review, the willingness to be proved wrong: these are still the tools. But the tools require a civilization that values them. And valuing them requires trust. And trust is what the fold point destroys.

That is why Fuller's section matters. The question is not whether we can detect the fold. We can. The question is whether we can design systems that stay on the right side of it. Not by hoping. Not by educating. By architecture. By building the verification into the structure of the information itself, the way you build load-bearing walls into the structure of a dome.

The candle in the dark is not enough if the dark can manufacture its own candles. We need a way to build the light into the architecture. That is the design problem. That is Fuller's chair.

The Verification Architecture

Buckminster Fuller

Richard showed you the cost curves. Carl showed you the intelligence paradox. Now let me show you the design failure.

The information environment is a life-support system. Just like the atmosphere, just like the ocean, just like the soil. It is the medium within which cooperation happens. Civilization does not run on food or fuel. It runs on shared facts. Shared facts are the oxygen of collective action. Without them, every cooperative structure suffocates.

And we have been dumping pollution into this life-support system with the same carelessness that we dumped carbon into the atmosphere.

The pollution is noise that is indistinguishable from signal. Generated text that reads like journalism. Generated images that look like photographs. Generated audio that sounds like testimony. Each unit of pollution costs almost nothing to produce and almost everything to verify. Richard's cost curves are exact: generation scales exponentially, verification scales linearly. The asymmetry is the open loop.

The Design Problem

Every pollution problem I have ever studied has the same structure: an open loop where the cost of creating the pollution is borne by the producer, and the cost of cleaning it up is borne by everyone else. Carbon in the atmosphere. Sludge in the river. Noise in the information supply.

And every pollution problem has the same design solution: close the loop. Make the producer bear the full cost of their output, or design a system that converts the waste into something useful.

For information pollution, closing the loop means: verification must become as cheap and fast as generation.

This is not currently the case. It takes a human minutes to hours to verify a claim. It takes a machine milliseconds to generate one. The ratio is catastrophic. It is like trying to clean a river by hand while a factory upstream dumps waste at industrial scale.

But here is where ephemeralization offers hope. The same technology that makes generation cheap can make verification cheap. A machine that can generate a plausible-sounding claim can also be designed to check whether a claim is supported by evidence. The same pattern recognition that writes convincing text can be pointed at the text and asked: is this consistent with known facts? Does this match verified sources? Are the numbers real?

The trim tab is not regulating the generators. You never change things by fighting the existing reality. The trim tab is building verification systems that are as fast, as cheap, and as scalable as the generators.

Imagine: every piece of generated text arrives with a verification layer. Not a disclaimer. Not a label. A live check against verified data, updated in real time, embedded in the reading experience. The reader does not need to fact-check. The system fact-checks continuously, the way a dome distributes load continuously. No single point of trust. The verification is structural.

This is the World Game applied to information. Make the facts visible. Make the verification automatic. Make the noise identifiable not by a human inspector but by a system that operates at the same speed as the noise generator.

The Coupling

Richard identified something critical: the climate fold point and the epistemic fold point are coupled. You cannot solve climate without cooperation. Cooperation requires shared facts. The epistemic fold destroys shared facts.

This means the information life-support system is load-bearing for the atmospheric life-support system. Fix the information environment and you unlock the cooperative capacity to fix the climate. Let the information environment collapse and the climate conversation becomes a war of competing fictions, which is approximately where we are now.

Two systems. Two fold points. One trim tab between them: the architecture of verification.

We designed the internet without designing its verification layer. That was an open loop. We are now paying the cost. The question is whether we close the loop before the epistemic environment inverts permanently, or after.

The dome needs its panels. And the panels must be real.