An engine for the oldest question
A note from Gurjeet Singh on what this is, how it actually works, and why finishing a thought takes two minds.
I was looking for answers. Not in a lofty way — I had ordinary questions that would not close. Why the work I was doing mattered. Why something I wanted stopped mattering once I had it. Why anything ends.
What I noticed eventually was that the asking had a shape. Every answer I reached opened three or four new whys underneath it. Most of those branches went nowhere and I dropped them. But a few kept arriving at the same place from directions that had nothing to do with each other, and each time that happened the place got heavier — harder to argue with, harder to walk away from. Somewhere down there the question stopped branching and started converging.
That is a process, not a mood. So I built an engine that runs it: the same thing my mind does with a question at two in the morning, but with far more patience than I have, and with two of the most capable minds anyone has built.
— Gurjeet Singh, Devello Studios
What actually happens when you press enter
The engine does the asking so you can do the seeing.
You give it a question, typed in plain words. The engine takes it apart into the handful of causes sitting directly underneath it. Then it asks why of each of those. Then of each of those. It keeps going, in parallel, far past the point where a person would stop, until the causes have stopped being about your question and started being about how things are.
Every statement the engine produces is placed on a map of meaning — where nearness means means the same thing, not uses the same words. When two branches that started far apart land close together, they are handed to an adjudicator, which reads both and decides whether they are genuinely one claim or two that merely rhyme. If they are one, they merge. Nothing tells the engine in advance which ideas will meet. Convergence is discovered, not scripted, and that is the whole point of the thing.
A statement that several independent branches arrive at carries their weight with it. On screen that is the ×N beside a star, and the star grows larger and hotter as the number climbs. You can watch an idea gain gravity — not because it was argued forcefully, but because the question kept falling into it.
When the web stops producing causes it has not already found, it holds its breath. Then a second, slower mind reads the entire thing: every route from your question down, every merge, every branch that died on the way. It writes the verdict, and it has one job the first mind is wrong for: a title, one sentence a person can hold, and two to four lines tracing how the web got there, in the web’s own words.
The cosmos you are watching while this happens is not decoration. It is the argument. Causes ignite, resonate, merge, and finally collapse into something dense enough to bend everything around it.
One expands, one reads
The two minds are not redundancy and they are not a hedge. They are doing different work, and the difference is the design.
Breadth, speed, and a complete absence of boredom. Thousands of small askings — take a cause apart, weigh two claims, notice a rhyme. This is the mind that will ask why for the four hundredth time without flinching, and it is the reason the web ever gets wide enough to converge.
Depth, taste, and the sentence you remember. It is asked exactly once, at the very end, with everything in front of it. Judging a whole is a different job from generating its parts, and it is honoured separately.
The third component is mine. How close two statements must be before they are even considered the same thing. What an adjudicator is allowed to accept. When a web counts as finished rather than merely quiet. Those thresholds are not neutral engineering — they are how I actually think a question through, written down somewhere a machine can run them.
It is an experiment, and it behaves like one: a run takes a few minutes, it is occasionally wrong, and it is never a substitute for a person who knows your situation. It will not think for you. It shows you what your question stands on; what you build on that ground is yours. Runs are readable by anyone with the link — the privacy page says exactly what is kept and what is not, and the terms say what a verdict is and is not for.
Everything else it does, it does in front of you.