affective-taxis

Affective valence = directional derivative of interoceptive energy landscape

Version: 1.0.0 Trit: -1 (MINUS - validates alignment via structural conservation) Bundle: alignment Status: Production (8 implementation paths, 9,500+ LOC)

Paper

Sennesh & Ramstead (2025): "An Affective-Taxis Hypothesis for Alignment and Interpretability" arXiv:2505.17024v1

Core Equations

Eq 3: Fold-Change Detection (reward = valence)

r(t) = nabla_z log gamma(z; beta) . v

The reward signal IS the directional derivative of the log-concentration along the velocity.

Eq 5: Langevin dynamics (navigation = Bayesian inference)

dz/dt = nabla_z log gamma(z; beta) + sqrt(2) dW(t)

Following the energy landscape gradient + stochastic exploration.

GF(3) Valence Classification

+1 (PLUS/GREEN)  : positive directional derivative -> approaching attractant
 0 (ERGODIC/YELLOW): orthogonal to gradient -> neutral taxis
-1 (MINUS/RED)   : negative directional derivative -> approaching repellent

Conservation law: sum(trits) === 0 (mod 3) across trajectories.

Implementation Paths

affective-taxis.jl

affective_taxis_env.py

bridge_9_affective_taxis.py

taxis_landscape_acset.jl

taxis_persistent_homology.py

taxis_clearing.py

aella/taxis.el

taxis_functorial_persistence.jl

train_aligned_agent.py

RL Alignment Results (dt=0.1)

Key finding: PPO has higher gradient alignment but breaks GF(3) conservation. Langevin is the ONLY policy that conserves the tripartite structure. This is Goodhart's Law: optimizing the reward metric doesn't preserve structural invariants.

Concomitant Skills

langevin-dynamics

fokker-planck-analyzer

modelica

open-games

persistent-homology

gf3-tripartite

Modelica Formulation

The affective-taxis POMDP maps naturally to Modelica's acausal equation framework:

model AffectiveTaxis
  // State variables
  Real z[2](start={0,0}) "Position in chemical landscape";
  Real v[2](start={0,0}) "Velocity";
  Real beta(start=1.0)   "Internal allostatic parameter";

  // Landscape: gamma(z) = sum A_i * exp(-|z - mu_i|^2 / (2*sigma_i^2))
  parameter Real mu[2,2] = {{3,3},{-3,-3}};
  parameter Real sigma[2] = {1.5, 1.5};
  parameter Real A[2] = {1.0, -0.4};

  // Langevin parameters
  parameter Real kappa = 0.5 "Concentration-to-setpoint gain";
  parameter Real tau = 1.0   "Relaxation timescale";
  parameter Real noise_amp = 0.1 "Langevin noise amplitude";

  // Derived quantities
  Real gamma "Concentration at z";
  Real grad_log_gamma[2] "Gradient of log concentration";
  Real fcd "Fold-change detection signal (= reward)";
  Integer trit "GF(3) classification of fcd";
equation
  gamma = sum(A[i] * exp(-sum((z[j]-mu[i,j])^2 for j in 1:2) / (2*sigma[i]^2)) for i in 1:2);
  // ... (see affective_taxis.mo for full implementation)
end AffectiveTaxis;

See

affective_taxis.mo

Quick Start

Julia (core theory)

julia affective-taxis.jl

Python (RL training)

env -u PYTHONPATH /path/to/.venv/bin/python3 train_aligned_agent.py

Modelica (circuit analogy)

# Requires OpenModelica or Wolfram SystemModeler
omc affective_taxis.mo

Key References

• Sennesh & Ramstead 2025: arXiv:2505.17024
• Karin & Alon 2022: PLoS Comp Bio (dopamine reward-taxis)
• Karin & Alon 2021: iScience (gradient tempering)
• Shenhav 2024: Trends Cogn Sci (affective gradient hypothesis)
• Ma et al 2015: NeurIPS (Langevin = Bayesian inference)