To be slightly more clear, it looks like they're saying that there are two _types_ of neurons, not two physical neuronal cells.
From the article's introduction:
"We demonstrate that the core decision-making system
consists of just two neuron types—a phasically firing
command-like excitatory neuron, encoding the presence of
food, and a tonically firing modulatory neuron, acting as a gain controller for the animal’s motivational state—and we
characterize these two pathways and their interactions at the
point that a decision is made."
Almost 25 years ago, Jonathan Connell's book "Minimalist Mobile Robotics(1)" briefly discusses the behavior network of the pond snail, and maps it using Brooks' Subsumption Architecture. The book then goes on to discuss the implementation of a robot that roams a MIT building, seeking empty soda cans and returning them -- all without centralized control, mapping, or data collaboration between a number of distributed processors. It's fascinating!
But getting back to the snail being discussed: it has a very simple map of behaviors.
If wet, seeks light and crawls upwards. If horizontal, it seeks dark. If vertical it crawls up. If upside, it seeks light. When dry, it stops to feed. (Or something like this... I'm going off of memory from 20 years ago.) These very simple behaviors can lead this snail to crawl out of the water along a shore line and eat at the vegetation that exists at the boundary between water and air.
It is quite amazing to see it all reduced to a handful of behaviors which mask lower-level behaviors in a diagram which can be sketched on a cocktail napkin.
From the article's introduction:
"We demonstrate that the core decision-making system consists of just two neuron types—a phasically firing command-like excitatory neuron, encoding the presence of food, and a tonically firing modulatory neuron, acting as a gain controller for the animal’s motivational state—and we characterize these two pathways and their interactions at the point that a decision is made."