A principle in isolation describes a slice

Studying a principle alone tells you what one variable does when everything else is held still. Real behavior is rarely so tidy. Consider a single choice: a smaller reward now, or a larger one later. Its value depends on the amount, the delay, the probability, the effort to obtain it, whether it is a gain or a loss, and the budget the organism is working within, all at the same time. Study any one of these alone and you describe a slice. Study how two of them interact and you begin to describe behavior.

The combination landscape

To see the whole problem at once, it helps to lay it out. Below is every behavioral process paired with every other. There is, tellingly, no settled list of these principles to begin from. Miltenberger and Flores (2024) surveyed 144 behavior analysts and found consensus on only 16 operant and 5 respondent terms, which is close to the set used here.

A cell is shaded where the field has co-studied a pair, highlighted where our lab has taken one on, split diagonally where both are true (i.e., our work in the upper triangle, the field's count in the lower), and left dark where no one has. Counts come from 46,988 articles published between 1956 and 2026 across 22 behavior-analytic journals (e.g., JEAB, Behavioural Processes, Learning & Behavior, JABA), scanned for the vocabulary of each process. Hover for a count; click any cell to read the studies, or to propose an open one with us.

26 processes shown. Counts are a lower bound: a study that arranges a process without naming it in its title or abstract is not counted here, and the process list itself remains a working draft.

Two things stand out. First, the field has worked densely in one corner and thinly nearly everywhere else. Pairings amongst the classic operant and respondent processes (e.g., stimulus control with discrimination, extinction with positive reinforcement) carry a median of six papers per cell, and the densest cell holds 318. Second, and more telling, the cells that pair one of those classic processes with a decision dimension (e.g., delay discounting, response effort, economic context) carry a median of one. Of the 325 pairings on the board, 72 have never been co-studied at all, and 186 rest on fewer than five papers.

Stated succinctly, the combination space is less empty than it is thin, and it is thinnest exactly where the two halves of the board meet. How a reinforcement history interacts with how an organism discounts a delayed outcome is, at present, mostly an open question. That seam is where this lab works.

What our lab has taken on

Our combination studies start in the value of choice. We have crossed delay with probability, effort with probability, amount with delay, and gains with losses across both. We have asked how the economic context an organism sits in (its earnings budget) reshapes those functions, and how verbal stimuli move risky choice. Each study adds one cell, and each cell narrows the gap between what we can model in isolation and how behavior actually allocates.

Scaling with AI and machine learning

Cell by cell is slow, and the space is vast. So the second half of this line moves from studying combinations one at a time to modeling them at scale. We took the generalized matching law from an N of one to an N of a million (Cox et al., 2022), used natural language processing to read behavior from text, and integrated reinforcement-learning models with behavior science to predict the next response (Cox & Santos, 2025). The aim is to let computational and AI methods explore the combination space faster than hand-run experiments can, and to feed what they find back into the lab. This is where this line meets the artificial-organisms work.