Note: All names used in Chiron are fictitious.
On a Friday evening in early spring, two clinic owners found themselves sitting across from each other at a bar not far from the conference hotel. Both had just finished a week dealing with staffing problems, insurance calls, and the endless puzzle of running an applied behavior analysis practice.
Elena, a PhD and BCBA-D owned a mid-sized clinic outside Denver.
Thomas, an MS and BCBA ran a fast-growing ABA company in Arizona.
The first round of drinks arrived.
Thomas leaned back in his chair and smiled.
“Honestly,” he said, “this year feels different. AI is finally fixing the boring parts of this job.”
Elena raised an eyebrow.
“Oh?”
“Session notes, treatment plan drafts, even summarizing assessment data. My team drops everything into an LLM and it writes clean reports in seconds.”
He took a sip of his drink.
“Our productivity is up almost thirty percent.”
Elena nodded slowly.
“Thirty percent is impressive,” she said.
Then she asked a question.
“How often are you checking what it produces?”
Thomas shrugged.
“It’s usually right.”
Statistical Fluency vs. Clinical Reasoning in Behavior Analysis
Large language models (LLMs) are remarkable tools (Behind the Wrapper: What You Need to Know About LLM-Powered Tools). They produce fluent text, summarize information quickly, and can accelerate many administrative tasks in clinical practice. But fluency is not the same thing as understanding. At their core, LLMs do something very specific: They predict what an average human might say next in a sequence of words.
They do not observe behavior.
They do not analyze contingencies.
They do not evaluate function in context.
They predict language patterns. This distinction matters for behavior analysts.
Applied behavior analysis is a contextual science. The meaning of behavior emerges from relations between the organism and the environment. Functional relations are discovered through observation, measurement, and analysis of contingencies in context.
LLMs operate on an entirely different foundation. They analyze textual patterns (i.e., topographical relations), not environmental contingencies. This means that an LLM can sound clinically intelligent while being functionally wrong.
