Note: All names used in Chiron are fictitious. Additionally, this is the first of eight episodes in which we build a story arc using the same characters. At the end, you will find a character cheat sheet to help keep everyone straight from episode to episode.
The Patterns Beneath the Interface
The conference hotel overlooked the mountains just enough to make everyone feel reflective after the final keynote ended. The attendees drifted toward the hotel bar downstairs where clusters of clinicians compared payer frustrations, staffing problems, and whatever new software platform their organizations had recently purchased.
Rowan (a senior BCBA) sat near the corner reviewing supervision notes on his laptop while Mira (a green BCBA) leaned across the table scrolling through a new AI-assisted documentation system.
“You have to admit this is getting ridiculous,” she said.
Rowan barely looked up. “Ridiculous good or ridiculous bad?”
“Both.” Mira turned the screen toward him. “Watch this.”
The platform generated a session summary in less than three seconds. Target progression prompts were sent by the system for the KPIs the clinicians valued most, caregiver participation, treatment integrity concerns, and even recommendations for what to include in follow-up supervision.
Rowan read silently for a moment.
Then he laughed.
“This thing is starting to sound exactly like me.”
Modern AI systems often feel more mysterious than they really are. Most of the systems clinicians are now encountering are not reasoning in the human sense. They are mathematical prediction systems trained on enormous amounts of human-generated examples. Large language models, recommendation engines, and adaptive documentation tools all work through the same general process: identifying statistical regularities in data and generating outputs that are likely to be reinforced as useful by future users.
That distinction really matters because prediction systems do not need consciousness to influence professional behavior. A calculator changes how people solve math problems without understanding mathematics. GPS systems change how people navigate cities without understanding geography. AI systems increasingly shape clinical workflows the same way. Not by “thinking” like clinicians, but by predicting which outputs clinicians are most likely to accept, edit, approve, or repeat.
From a behavior analytic perspective, this is important because the environment does not need intention to exert control. Contexts shape behavior through contingencies whether the shaping system is another person, an organization, or a digital platform. The more often a clinician interacts with an AI system, corrects it, reinforces it, or relies on it, the more that interaction itself becomes part of the behavioral environment influencing future responding.
The conference hotel bar had become crowded as the night went on. Small clusters of clinicians drifted between conversations, carrying drinks, conference tote bags, and exhausted expressions that only appeared after twelve straight hours of talking about insurance systems and human behavior.
Someone near the back argued about caregiver training requirements.
Another group debated whether AI-generated session summaries would eventually become billable documentation.
Mira refreshed the screen again.
“Look at this,” she said. “It even prioritized the supervision concerns in the same order I would’ve.”
Rowan leaned closer, still half-smiling.
“That’s either impressive or deeply concerning.”
Juniper (clinical lead) heard the comment from the next table over and glanced toward the screen. She smiled politely, but the joke lingered with her longer than she expected. He pulled out a chair for Juniper and set his phone beside his drink. The screen still displayed unanswered messages from scheduling staff.
Earlier that week, the same platform had generated supervision language nearly identical to phrasing she personally used during difficult clinical reviews. Not generic ABA terminology. Specific cadence. Specific tone. The same careful balance she used when trying to soften corrective feedback without diluting clinical urgency.
At first, she had been impressed.
Then she realized the system was not simply helping clinicians write.
It was learning how clinicians thought.
Most people still imagine artificial intelligence (AI) as something futuristic or mechanical, but modern AI systems are often much simpler and much stranger than the public imagines. Most current systems are not thinking in the human sense of the word. They are prediction systems. Large language models (LLMs) work by identifying statistical relationships across enormous datasets and then predicting the most probable next output given the context surrounding it.
In practice, this means every correction clinicians make becomes valuable. Every rewritten sentence, modified recommendation, supervision edit, or treatment adjustment teaches the system something about what humans inside that environment consider useful, appropriate, or professionally correct.
