The Fly For Congress

About

Is any of this real?

Some of it. Here is exactly which parts.

Real

  • The brain. 223 neurons and 3,420 measured synaptic connections, with real body IDs, from Janelia’s NeuPrint male-cns:v1.0 — a published wiring diagram of a male fruit fly’s central nervous system.
  • The votes. Congressional roll calls from 1965 to 2014 that carry a Peltzman issue code — about 46,000 of the 57,000 eligible — from Voteview.
  • The map of Congress. DW-NOMINATE coordinates for every member who cast those votes.
  • The circuit really does settle under a stimulus, and it really does turn left or right.

Invented

  • The mapping from a policy issue to a sensory stimulus. We made it up.
  • Reading “left minus right” as a political stance.
  • The fly, the suit, the biography, and the entire premise.
  • Any suggestion that a fly can hold office. It cannot. That is the joke.

How a fly takes a position

Each of thirteen issue categories has a direction in ideological space: the average position of the members who voted yes minus the members who voted no. That direction is encoded as a lateralized stimulus on a measured locomotor circuit of 223 neurons. The circuit settles for about three simulated seconds and the stance is read off the left–right asymmetry of its descending motor output: the circuit literally turns toward one side or the other. Forward drive is read separately, as appetite to act.

How a fly changes his mind

With his own learning circuit. The mushroom body is where a fly forms associations — Kenyon cells carry a sparse code, dopamine neurons teach, and output neurons bias approach or avoidance. We exported it from the same connectome: 589 of 4,064 Kenyon cells, all 97 output neurons, 332 dopamine neurons and 25,313 measured connections. Each side of each issue is presented to the Kenyon cells as an odor would be. A persuasive argument is a training trial: the argued side is paired with appetitive dopamine, and the synapses from the active Kenyon cells onto avoidance-promoting output neurons are depressed — the mechanism the mushroom body is known for. Which output neurons promote approach and which avoidance is read from the measured dopamine wiring, following Aso et al. 2014, and the canonical cases check out.

The measured synapse counts are never modified. Learning is a bounded gain on each synapse that can fall to a fifth of measured and relaxes back over about a month. One conversation is a nudge; a reversal takes roughly forty-five. And because nearby issues share Kenyon cells, moving him on one thing moves him a little on its neighbours — the same generalization a fly shows between similar odors.

What we do not claim

  • This is not a model of political cognition, decision-making, or belief. Approach and avoidance are the fly’s; what is approached is our choice.
  • A different issue-to-stimulus mapping would produce different positions.
  • DW-NOMINATE is a spatial summary of voting behavior, not a measure of belief.
  • Issue coding ends in 2014. The candidate has nothing to say about anything after it.
  • Five of thirteen categories lack sufficient recorded vote outcomes and use a documented neutral substitution rather than a guess.

Method and measurements are documented in the project’s ENCODING.md. Connectome: NeuPrint. Votes: Voteview.