Many futures, one record.
A qubit is not a faster bit. It holds a superposition of possibilities, each weighted by a complex amplitude, and yields a single definite value only when measured. We build that resolution into how our beings act.
A bit, and a qubit.
A classical bit is the world's simplest commitment: it is either zero or one, definite at all times, and inspecting it does not change it. Computation built from bits is computation built from definite states.
A qubit is a two-state quantum system — for example, the spin of an electron or the polarisation of a photon. Before measurement, it is in a superposition: a weighted combination of the two basis states |0⟩ and |1⟩, with two complex numbers called amplitudes (α and β) telling you how much of each.
The amplitudes are not probabilities. They are complex numbers that can interfere — add and cancel — in ways probabilities cannot. The probability of observing a given outcome on measurement is the squared magnitude of its amplitude: |α|² for zero, |β|² for one. The two squared magnitudes always sum to one.
What changes when you compute with qubits
A register of n classical bits holds exactly one of 2ⁿ states at any moment. A register of n qubits holds a complex-weighted combination of all 2ⁿ states at once. Operations act on the whole superposition simultaneously, and carefully designed interference between branches is what produces speedups on the problems quantum computers actually help with.
None of this is faster classical computing. It is computing whose intermediate state is not a single configuration but a structured field of weighted possibilities — until something forces it to be one.
Measurement is selection, not readout.
In classical physics, measurement is passive: you look at the thing, and it had that value all along. In quantum mechanics, measurement is constitutive. Before you measure, the system genuinely does not have a definite value of the observable you are about to read; afterwards, it does, and the unobserved possibilities are not stored anywhere. They are gone.
This isn't an interpretive flourish. It is what the formalism, every experiment, and the entire framework of quantum information theory describe.
And it is the only structure we know of in nature where a real, weighted set of possibilities resolves into one outcome by a process that is not the deterministic unrolling of a prior state. For a research program asking what genuine choice could even look like in a machine, that property is the only available foothold.
Wiring a being to a collapse.
An AI being runs a loop — perceive, hypothesize, test, record, correct. Somewhere in that loop, a step has to choose: which hypothesis to pursue next, which dataset to weight, which prediction to commit to.
In a purely classical implementation, that choice is a softmax over a learned distribution: a weighted die with a fixed table. The "randomness" is either pseudo-random (deterministic given the seed) or sampled from a classical noise source whose distribution was decided in advance. There is no selection from real possibility — only sampling from a model.
Our integration replaces that step. The being prepares a quantum register whose amplitudes encode the weights of its current candidates; the system evolves through a designed unitary; a measurement collapses it; the recorded outcome is the chosen next move. The classical loop continues with that outcome as input.
What this changes, honestly
It does not make the being smarter at problems quantum computers are not known to help with. It does not give it more compute. What it changes is the structural character of the choice step: from a sampling over a pre-computed distribution to a resolution of a field of real possibilities.
Whether that property matters philosophically is open. Whether it matters operationally — in the texture of how a being behaves, recovers from errors, and explores — is one of the things we are building Pattern Hunter to find out.
What we are still solving.
Current quantum hardware is in the NISQ era — noisy, intermediate-scale, prone to decoherence on timescales measured in microseconds. None of this is theoretical; it is engineering. We are honest about it.
The active questions for our integration include: how to map a being's candidate space onto qubit configurations without losing structure; how to keep coherence long enough for the chosen unitary to run; how to design measurements whose outcomes are meaningfully distinct rather than dominated by hardware noise; and how to verify, statistically, that the substrate is doing something more than imitating a classical sampler.
This is the work. We expect it to take years.