Magical Thinking
Fig. 0 — The mark under construction
rₙ₊₁ = rₙ / φ—

Magical Thinking

§ 01 — Prima materiaProposition

That which is below is like that which is above.Tabula Smaragdina

A research laboratory building living, instrumented models of human neural function — to understand how experience becomes learning, and to turn that understanding into better medicines and better artificial intelligence.

Our founding hypothesis is that the most useful properties of neural systems emerge from the interaction of development, cellular diversity, circuit architecture and closed-loop experience — and that these properties are measured more faithfully through controlled perturbation than through observation alone.

We combine human neural cultures, organoids and assembloids with chronic bioelectronic interfaces, controlled environments and AI-guided experimentation, to map how circuits change, remember, recover and fail.

We begin with a concrete test: whether reproducible, challenged neural systems reveal drug effects that simpler measurements miss. From that foundation — durable biological learning, adaptive electronic connections, and mechanisms that improve learning in silicon.

The universe assembled a machine that can wonder about the universe. We would like to see the wiring.— Operating thesis
§ 02 — NigredoFive questions
at the edge of mind

The obscure, by the more obscure.Alchemical maxim

Probe tip, pens and the traces they drawTwo identical cultures, one challengedSame tissue, same recorder, same analysis. The challenged culture answers the probe with an evoked response that recovers in about 2.5 s; the resting one shows only baseline. The readout is the largest swing on each chart. H1 asks whether that extra information improves a held-out drug-response decision.
  1. H1

    Function under challenge carries useful information.

    ClaimEvoked, recovery and state-transition measurements improve held-out drug-response discrimination beyond rest-only metrics.

    Fails ifThe added measurements are less repeatable, leak labels, or do not improve a prespecified held-out decision.

  2. H2

    Better access is valuable only when it reveals more.

    ClaimConformal or embedded 3D interfaces increase stable, spatially discriminable information without unacceptable tissue injury.

    Fails ifExtra channels are redundant, unstable, or bought at the expense of viable tissue.

  3. H3

    AI can buy information, not just generate plans.

    ClaimUncertainty-aware experiment selection reaches an externally checked conclusion with fewer accepted units than a strong fixed design.

    Fails ifThe efficiency comes from excluding difficult batches, optimizing a proxy, or moving the criterion.

  4. H4

    Experience can change the biological substrate.

    ClaimContingent experience produces retained improvement after training ends, with encoder and decoder frozen.

    Fails ifGains are explained by drift, artifacts, external model adaptation, or recent stimulation alone.

  5. H5

    A discovered mechanism can transfer out of biology.

    ClaimA mechanism identified by causal perturbation improves an artificial learner on held-out continual-learning tasks.

    Fails ifThe benefit disappears once compute, memory, tuning and supervision are matched.

§ 03 — AlbedoThe instrument

Learning never exhausts the mind.Leonardo da Vinci

Four elements, kept apart. Separately addressable, never conflated — every channel keeps its identity.

Electrical

Recording and stimulation through individually insulated sites.

Optical

Imaging and light-based readout, registered to the electrode map.

Fluidic

Perfusion as an engineering variable — not mistaken for vasculature.

Chemical

Controlled exposure and context, with viability tracked alongside.

The addressed module and its own lead; the others stay idleThe culture dish all four shareOne stage, four instruments. Each module reaches the dish through its own lead and its own port, so any reading can be traced to exactly one source.
The work we exist to do

The bioelectronic neuropil

A soft, porous scaffold threaded with individually insulated traces, every exposed site able to listen and to speak. Tissue grows into the lattice and through it, until living circuitry and electronics are no longer pressed against each other but woven into one architecture.

This is the instrument this lab was founded to build. We are certain it can be done, and we intend to be the ones who prove it.

MT—NP 01
The addressed site, its own trace, its own pinPorous scaffold, layered latticeShown with sixteen sites. Each exposed site keeps its own insulated path, so tissue grown through the lattice can be heard, and answered, one channel at a time.
§ 04 — CitrinitasThe spiral

Make haste slowly.Festina lente

The work proceeds as a spiral. Each turn rests on the one inside it and widens what can be asked. The innermost turn is a measurement we can trust. The outermost has no name yet.

  1. IMeasurement we can trust
  2. IIRest and challenge
  3. IIIGeometry and mechanism
  4. IVLearning in living tissue
  5. VExperiments that choose themselves
  6. VIMechanism into silicon
  7. VIISynthetic pathways
  8. VIIIElectronic temporal companions
§ 05 — ConiunctioCorrespondence

Know thyself.Inscription at Delphi

We are assembling scientific leadership, tissue and instrument partners, and those who fund patient science.

inquiries@magicalthinking.ai