Teaching the jar to learn

Right now the mouse's wiring is fixed. Every connection is exactly as the microscope found it. This page is where that will change, and where you will be able to watch it.

Reading the jar.

Where it is now

The jar runs on 1,627 neurons from one mouse's visual cortex and the 52,181 connections between them. When a stock moves, the signal travels through those connections to the cells that vote, and the votes decide the trade. What happens after the trade never reaches the brain. A trade that made money and a trade that lost money leave it exactly as it was.

How a trade will change it

While the mouse listens to a stock, the brain will keep a note of which connections carried the signal. A connection counts when the cell sending the signal fired just before the cell receiving it, which is the pattern that strengthens connections in real brains.

When the trade closes, the jar will know whether it made or lost money, and that result goes back to the connections in the note. A profitable trade makes them stronger, so the same kind of price move is more likely to produce the same trade next time. A losing trade makes them weaker. A bigger win or loss makes a bigger change, and no single trade can move a connection more than a fixed limit. Connections that had no part in the decision stay as they were.

A buy and its sell can be days apart. The note is saved at the moment of the buy and the result is applied when the sell happens, so the change lands on the connections that actually made the decision.

In a real brain, dopamine plays the part of the result: it arrives after the action and changes the connections that were recently active. The jar will do the same thing with money in place of dopamine.

Two brains

The original mouse will stay untouched. A copy will run next to it with learning switched on, trading the same prices with its own $10,000 book. The original shows what the real wiring does. The copy shows what that wiring turns into once trades can change it.

What this page will show

Each closed trade next to the connections it changed and by how much. How far the learning copy has moved from the real mouse, as the share of its connection strength that is no longer what the microscope found. And the two accounts side by side, so you can see whether learning makes the jar trade better or only differently.

What might go wrong

Each cell only has a few dozen connections inside the sample, and the jar makes a few trades a day. Learning will be slow, and with so few trades it can strengthen connections that were only lucky. If that is what happens, this page will say so.

Sources

1. Izhikevich, E. M. Solving the distal reward problem through linkage of STDP and dopamine signaling. Cerebral Cortex 17, 2443–2452 (2007). doi:10.1093/cercor/bhl152
2. The MICrONS Consortium. Functional connectomics spanning multiple areas of mouse visual cortex. Nature 640, 435–447 (2025). doi:10.1038/s41586-025-08790-w