I think there is a difference between emotions and neurotransmitters.
Serotonin might color a sandwich as being interesting, friendly and non-threatening, but I think there is a learned desire for food as a response to low blood sugar that has the capacity to inhibit non-food related next states.
So, I’m in a room, and a lot of possible next states of the “me in the room” model are somewhat lit up… I could open the window, sit down, flip the table, go and eat the sandwich.
There is a center in my brain that observes low sugar and wants to solve for that bad state with a desired next state “me fed”… it inhibits any of the next states of the world model that are not also associated with me being fed, so I go with the only one which is still lit up… “Eat the sandwich”… Once that actionable state is lit up, it reinforces itself while signaling sub-next-steps (walking, picking up a sandwich, eating) that are appropriate to the world model as it stands right now. Once the final step is reached, either it signals the “Eat the sandwich” plan to stop being lit up or my lack of hunger allows some other desirable next state to inhibit the currently executing plan…
All of the above would also work in a dog or a fish… I think where humans differ in part is in the depth of their call stacks… I might have a sub-plan for eating that includes placing a call to Domino’s Pizza and then a sub-plan for getting my wallet… and I can execute the stack without losing track of the original goal for some depth of stack.