How do you make a wavefunction collapse without an observer?
That's the question the AI on the other page is preparing to answer — and it's a 100-year-old question with a fresh twist. Part 1 below is the short history of asking what measurement actually is; Part 2 is a tour of what a wavefunction does, because that's where the answer might live.
Scroll. The picture on the right will follow.
scroll to step through · part 1 is history · part 2 is wavefunction tour
Quantum mechanics has two rules — and nobody knows when to switch.
Quantum mechanics has been the best-tested theory in the history of science for almost a century. Lasers, transistors, MRIs, the chip in your phone — all of it sits on QM. So you'd think the rules would be clean.
They're not. There are two rules, and they contradict each other:
Rule 1: smooth. When nothing is measuring it, the wavefunction evolves smoothly and predictably by an equation Schrödinger wrote in 1926. Two possibilities can coexist — "here" and "there", at once.
Rule 2: jumpy. When something measures it, the wavefunction suddenly snaps to one possibility. The "and" becomes an "or".
What counts as a measurement? Textbooks don't say. A pointer? A photographic plate? A cat? A person? Where does Rule 1 stop and Rule 2 start?
Born saw the wavefunction is a probability map.
Schrödinger's equation gives you a complex-valued function ψ at every point in space. In 1926, Max Born had the move that made QM usable: |ψ|² at a point tells you the probability of finding the particle there if you look.
Suddenly the wavefunction wasn't a thing — it was a prediction. The bright fringes in a double-slit experiment, the spectral lines of hydrogen, the spin of an electron — all of it came out right when you computed |ψ|².
Born got the Nobel for this in 1954, by which time it was the backbone of all of chemistry and most of solid-state physics.
Bohr said: the observer does the collapsing.
If Rule 1 is smooth evolution, and Rule 2 is a jump, then something has to trigger the jump. Bohr's answer, the Copenhagen interpretation, was: the trigger is observation. A "classical" measuring device interacts with the quantum system, and the wavefunction collapses.
It works in practice. It's what every physicist computes with. But notice what it doesn't say: what counts as a measuring device. Is an electron in a wire a measurement? A photon scattering off the system? A bacterium? A person?
Bohr's answer was essentially: don't ask. Treat the measurement apparatus as classical, the system as quantum, draw the line wherever it's convenient, get the right answer. It was a practical truce, not a theory.
Einstein hated this. "Does the moon exist when nobody is looking at it?" he asked. Bohr, basically: don't worry about it.
Schrödinger drew a cat to make the absurdity obvious.
Erwin Schrödinger, the man who wrote the smooth-evolution equation, wasn't convinced by Copenhagen either. In 1935 he offered a thought experiment to embarrass the rule.
Put a cat in a sealed box with a radioactive atom and a vial of poison. If the atom decays — a quantum event — the poison breaks, the cat dies. Don't open the box. According to Rule 1, the atom is in a superposition: decayed and not. So is the poison vial. So is the cat. The cat is alive and dead at the same time, until you look.
Schrödinger's point: obviously this is wrong. Cats are not in superpositions. There's something missing from the theory — some rule that ends the maybe before anyone looks. We just don't know what it is.
Ninety years later, the cat is still in the box.
The environment does some of the job. Not all.
From the 1970s onward, Zeh, Zurek, Joos and others noticed something important. A real cat in a real box is being constantly bombarded by air molecules, infrared photons, cosmic rays. Each of those interactions entangles the cat with the environment, and the cross-terms in the wavefunction — the interference between "alive" and "dead" — get washed out extremely fast.
The maths is clean, predictive, and confirmed by experiments on molecules of increasing size. We can see the loss of interference as objects get bigger. This is decoherence, and it's not controversial.
But — and this is the catch — decoherence explains why we can't see the superposition any more. It doesn't say it's gone. The wavefunction still contains both branches; it's just that the cross-terms have leaked into the environment too thoroughly to ever recombine.
Decoherence turns "alive AND dead" into "alive OR dead — but I can't tell which from any local measurement". It doesn't actually pick one.
Today there are two flavours of answer.
Interpretations say standard quantum mechanics is already complete; what we need is a different way of reading it.
Many Worlds — every branch is real, every outcome happens in its own universe; there's no collapse, just branching.
Bohmian mechanics — particles always have definite positions, guided by a pilot wave; the wavefunction is real, but so are the particles.
QBism — the wavefunction is in the observer's head, not in the world; it's a bookkeeping device for personal belief.
These all give the same predictions as standard QM, so no experiment can pick between them. They're philosophy, not physics.
Collapse models say something different: there is a real, physical collapse, and Schrödinger's equation is approximately right but wrong in a tiny way. They modify the equation with a small extra term that does almost nothing to a single electron but adds up to a real collapse for a cat.
GRW — every particle has a tiny random chance per second of getting "hit" by a localising kick.
CSL (Continuous Spontaneous Localisation) — the same idea but smoothed into a constant background random nudge.
Diósi-Penrose — collapse is driven by gravity: a superposition of two positions costs energy in its own gravitational field, and that energy debt forces a choice.
Collapse models make different predictions from standard QM. They predict tiny extra heating, faint X-ray emissions, measurable limits on superposition size. Experiments are actively ruling out parts of the parameter space. This is a live research frontier.
This project takes the collapse-model route.
If collapse is a real, physical process, then the maths of it should be discoverable. There's a finite space of ways to modify Schrödinger's equation while keeping all the things we already know to be true. That space hasn't been mapped — only a few points in it have been tried, by humans, over decades.
An AI can try thousands of points. A second AI can compare each proposal against the literature and the experimental bounds, and say "this is already ruled out", "this is a relabelling of GRW", or "this is genuinely new — simulate it".
To follow what the AI is searching for, you need a feel for what a wavefunction actually is. So let's look at one.
A wavefunction is a number at every point in space.
Not a real number — a complex number, with a size and a direction (a phase). The size² tells you the probability of finding the particle there. The phase is what lets wavefunctions interfere.
The simplest one is a wavepacket: a lump of amplitude that's smeared out over a region, with a carrier wave underneath. Watch the picture: the violet glow is |ψ|² — where you'd be likely to find the particle. The wiggle on top is the real part of ψ — the carrier.
A particle "has a definite position" only if its wavefunction is a sharp spike at one point. A spread-out packet is a particle that's genuinely not at one place.
Two wavefunctions add. That's superposition.
If a particle could be in state A or state B, in quantum mechanics it can be in A plus B. The wavefunction is literally the sum of two wavefunctions.
For Schrödinger's cat, the cat-state is "alive + dead" — and the plus is doing real work. It's not a metaphor. The mathematics treats the alive-cat ψ and the dead-cat ψ as vectors, and the cat-state is the sum.
Superposition is what makes quantum computers powerful: an N-qubit machine can hold 2^N branches at once and let them compute in parallel.
When the two paths overlap, you get interference.
Send a single electron at a barrier with two slits. The wavefunction splits, goes through both slits, and recombines on the far side. Where the two pieces add up in phase, you get a bright fringe — most likely to find the electron there. Where they add up out of phase, dark fringe — never find it there.
Done one electron at a time, the pattern still appears. Each electron must have gone through both slits. That's the thing that's hard to swallow — but it's the standard double-slit, and it's been done with atoms, molecules of 2000+ atoms, even small viruses.
If you peek at which slit the electron went through, the fringes vanish. The peek is itself a measurement — it collapses ψ. This is the measurement problem in miniature.
The chance of any outcome is |ψ|² at that outcome.
This is the Born rule, the one bit of QM that translates from wavefunction-land to experiment. Square the amplitude, get the probability.
Every prediction QM makes — the energy levels of hydrogen, the rate of laser emission, the shape of an orbital — comes out by squaring some amplitude. It's how you get from the smooth wavefunction to the discrete clicks of a detector.
A collapse rule has to preserve the Born rule. If a candidate recipe makes detection probabilities deviate from |ψ|², it's ruled out by sixty years of experiments.
Two particles can share one wavefunction.
Prepare two electrons so their spins are opposite, but you don't know which is which. Send one to Alice, one to Bob. They could be light-years apart. Whenever Alice measures up, Bob measures down, and vice versa. Always.
That's not surprising on its own — Alice could have a red ball, Bob a blue ball, in sealed envelopes. What's surprising is what John Bell proved in 1964: no theory of pre-arranged red and blue balls can match the QM predictions. The correlations between Alice and Bob, when they choose measurements at random, are too strong to come from any hidden agreement.
Experiments since the 1970s have repeatedly confirmed Bell's spookiness — most recently with all loopholes closed, by Hensen et al. in 2015 and the Nobel-winning Aspect/Zeilinger/Clauser experiments. The universe really is non-locally correlated.
A collapse rule has to preserve this. If a candidate kills entanglement before Bell can measure it, it's ruled out. And it must not let Alice signal Bob faster than light — also ruled out.
So: what if the wavefunction just collapses?
Watch the picture. A wavepacket is spreading along the line, the way Schrödinger's equation says it should. The amplitude is smeared out over a region — the particle is genuinely in many places.
Then — without anyone looking — it snaps. A single point is picked from the |ψ|² distribution. The wavefunction reshapes itself into a sharp spike around that point. The "and" becomes an "or", physically, in the world, on its own.
That is what a collapse rule is: a tiny extra term in Schrödinger's equation that does almost nothing for a single electron (so a double-slit still works) but adds up, across the billion billion particles in a cat, to a fast macroscopic snap (so the cat is alive or dead, never both).
The AI is hunting for recipes — exact mathematical forms for that extra term — that do this without breaking any of the things we already know: interference, the Born rule, Bell violations, no faster-than-light signalling. There aren't many ways. The space is small. That's why a search makes sense.
watch the AI hunt live →Now you know what the seven tests on the homepage mean. Wave maths still works for small stuff · tiny things stay quantum, big things go classical · no faster-than-light signalling · Bell's spooky pairs still spook · macroscopic things settle to one outcome · doesn't secretly heat up matter · reduces to known maths in a sensible limit. Each is a test a recipe has to pass.
Want the actual equations behind everything you just saw — the Schrödinger equation, the Itô SDE form of CSL, the gravitational-self-energy of Diósi-Penrose, the AI's scoring? See the Maths page →
Or read the longer-arc version of the bet — The Big Idea page →