“From 0 to 1” has become a cliché, but the questions behind it are still hard. Why is going from 0 to 1 so much harder than going from 1 to $\infty$? Real innovation also seems to need broad knowledge that can look like a distraction. And now that AI seems able to compute almost anything, it is fair to ask whether it can take this first step.
Information theory gives a simple and intuitive way to think about these questions.
In information theory, entropy measures how uncertain a system is. High entropy means disorder and unpredictability. Low entropy means order and clarity.
In any field, all the possible theories, technical routes and designs form a huge hypothesis space $\Omega$. Innovation is the process of gathering information to rule out the hypotheses that do not work.
So innovation is less a sudden “eureka” moment and more a steady process of reducing entropy, slowly finding order in the fog.
To see why 0→1 is so hard, we need to separate it from 1→$\infty$.
1→$\infty$ is an optimization game. Once the iPhone existed, the shape of the smartphone was settled, and after Amazon proved its model, everyone knew how e-commerce worked. The goal is simply to be faster or cheaper. This is small, step-by-step entropy reduction inside a known structure, and the results are steady and predictable.
0→1 is “crossing the river by feeling for the stones”. You do not just lack the answers. Often you lack even the words to describe the problem. You are moving through an unknown hypothesis space with endless possibilities. Finding a structure that works and lasts in this mess is hard, and it is a nonlinear jump that cannot be undone. It takes more than courage. It is like trying to build a house on moving sand.
Copycats often try to copy the technical path of a success, and they fail. They can clone the “1”, but they cannot inherit the process that produced it, including the scars.
Innovation goes from high entropy to low entropy. Failures and dropped hypotheses are useful data. Together they form a “shadow map” that shows exactly where the dead ends are and which limits cannot be crossed. This map gives the creator a gut sense of where not to go.
The copycat sees the destination but not the route. They get a low-uncertainty result without going through the work of reducing the uncertainty. When conditions change or the path has to go further, they are lost, because they do not have the map of the danger zones. What really protects the creator is less the code than the unwritten knowledge of what failed.
This is not praise of knowing everything. It follows from the structure of the problem.
A narrow specialist may follow a path that others already know is a dead end. Their hypothesis space $\Omega$ stays full of noise. Knowledge from other fields works as a filter. Physics sets energy limits, and engineering tells you what can be built. Breadth helps less by letting you know more and more by letting you rule things out faster.
But breadth needs depth. Innovation means finding the underlying logic that explains strange results. Deep knowledge compresses raw facts into intuition. When the real “1” shows up, even in a rough form, a person with depth can see its potential. Without depth, people dismiss a revolutionary idea just because it does not look like what they already know.
Interdisciplinary work feeds on “mutual information”, the moment when a principle from field A suddenly breaks a deadlock in field B.
Peter Thiel’s “secrets” are the vertical jumps of 0→1.
They are rare entropy reductions with high value. A secret is a truth that most people do not believe. In terms of probability, an event with a low prior probability carries a lot of information when it happens.
This is why innovation often goes against intuition. If everyone agreed that an idea would work, the certainty would already be high and the return would be small. People who really change things often have to be right and alone for a long time. The value comes from the jump from total uncertainty to clear proof.
AI is a strong catalyst. It can clear obvious obstacles and find patterns across distant fields. It is very good at turning ideas into things, such as code, proofs or figures, and it saves us from routine work.
But it has a basic limit. It works on probability, not on perception. AI fits the curve of the majority, so it improves what already exists and does not design against the consensus. It gives the “average” right answer, while 0→1 needs the “exceptional” right answer.
Most important, AI is not accountable. It cannot believe in a “1” worth fighting for or bet on a future with no data. That risk, the cost of being wrong, is a burden only humans carry, and it is what drives real creation.
0→1 turns complexity into clarity, and it needs both breadth and depth.
Today AI frees us from much of the work of search and optimization. But as the search space shrinks, the qualities that cannot be automated, vision and grit, become what we lack most.
AI can light the lamp, but humans must decide where to walk in the dark. We can do this because we can judge what is valuable, and we are willing to take the risk that comes with that judgment.
Innovation is not magic. It is long, hard exploration. Information theory gives us a map, but we still have to walk the ground.