New behaviour, old systems (AURA Stories 3)
Behaviour analysts in search of a new procedure in the world of automation, where AI systems cannot create what does not yet exist
A book is not a shoe
Sofia looks at Yen sitting across from her. He is staring at her, tracking the smallest shift in her eyes, waiting to see where her gaze will fall so he can pick up one of the three objects on the table. She recognises the pattern immediately. Another instance of Yen latching on to yet another cue that should be irrelevant in auditory visual object selection, what most people would call receptive labels. Damn. She had thought, for a moment, that she had finally established a discrimination. But it was an illusion. Yen will pick up on any antecedent in the environment except the verbal one.
She already knows what tonight will look like. Another long stretch poring over the literature on establishing listener discriminations for common objects. By now she can recite most procedures verbatim. She has tried everything: conditional only, simple conditional, identity matching with a delay, differentiating tone of voice, using onomatopoeic sounds instead of words, strengthening the preceding observing responses, every permutation of most to least and least to most prompting, errorless learning and stimulus control transfer, progressive and constant prompt delays, even the old No No Prompt and extinction with differential reinforcement. None of it shifts control to the spoken word. Verbal stimuli of any kind simply do not evoke differentiated responding in any context: naturalistic, structured, daily life. She is at her wits’ end.
She looks at Yen’s serene face. He is sitting beautifully, calm, his hands on the table. She tries one more time. He lets her know he is ready for her to begin teaching by orienting his gaze to her face.
“Shoe,” she says, keeping her entire body still, without dropping her eyes to one of the objects.
Yen reaches for the book.
“Okay,” she mutters, “he is back to picking up the item in the centre.”
She exhales, long and slow, feeling the familiar mix of frustration. It is not his fault; if there is a mistake, it lies with her, the teacher. The student is always right, Fred Keller would say, as quoted by Sage. She searches her memory. What else had Sage said in the summer?
She reaches for her notebook of Sage’s “quotes of wisdom”, the one she started writing in by hand on that first day, when she realised just how useless her iPad was in this place. She flips through the pages until she finds it. “Teaching means intentionally arranging the environment so behaviour that is not there yet can appear, and eventually be brought under stimulus control. We do not wait or hope for it. We build it.” Sofia rolls her eyes at the text.
Fine. Beautiful words. But what do I need to do here? She closes the notebook. She needs to talk to Sage tomorrow.
New cue, old cue. New cue, delay, reinforce.
Sofia gets in before everyone else the following morning. She is sitting on the floor, surrounded by manuals and photocopied articles spread out in a wide circle around her. She thinks of AURA, which would have analysed the videos, cross referenced every published study on listener discrimination, and generated a neat sequence of steps for the technician. All she would have needed to do was press “approve”. It is rare for her to miss AURA’s efficiency, but today she does.
She feels the frustration rising, the kind that makes her long for someone to just tell her what to do. But Sage never works that way. Sage will not simply hand over a solution. She will make her work, think, analyse, break the contingency apart, and build it back piece by piece. Sofia knows this, as she hears:
“Is Yen’s listener discrimination keeping you up at night?”
Sage’s voice comes from behind her, warm, almost amused, as she hands Sofia a cup of coffee.
“That is a very good thing.”
Sometimes Sofia just does not get her.
“Why is that a good thing?”
“It means you are engaging with the problem,” Sage replies, settling onto the floor beside her. “You know there is a solution, but it is not immediately available. Its temporary absence forces you to keep looking, to keep searching.”
“But I have looked,” Sofia says, exasperated. “I have looked at all the available evidence, all the literature.”
Sage gives her a sideways glance.
“Does evidence based necessarily mean published?”
Sofia frowns.
“Then how does one come up with something new? Are we supposed to be bound solely by what has been done before? What happens when none of the procedures fit the problem or the learner?”
Sage gestures toward the sea of papers around them.
“Sofia, where do we start?”
Sofia sighs. Here we go… and stops herself from eye-rolling.
“We start by defining the controlling variables. Then we choose a starting point, what the child can already do.”
“There you go,” Sage says softly. “Exactly.”
She stands and dusts off her hands.
“But on this one, I will help you. Come on. Let’s go see Yen.”
Yen is lying on the carpet, meticulously lining up cars with mathematical precision in an L shape. If she took out a square ruler, Sofia suspects it would be a perfect ninety degree angle. He is a quiet boy, using a few gestures and pictures to communicate. She has never heard him produce a sound. Yen is excellent at imitating. She discovered that during yet another procedural attempt at establishing listener conditional discriminations. When she pointed to each item to ensure he was scanning the field, Yen got up, walked next to her, and pointed to each object in the same sequence, touching the exact spot she had.
Sage watches him for a moment.
“What can he do already?”
“He can match with different topographies: pointing, giving, and finding identical items around the room,” Sofia replies. “If you show him something, he will find the same thing, as long as the sample stays present. I tried identity matching with a delay, but the moment I removed the sample, he kept taking my hand to show him the object.”
“Okay,” Sage says, nodding. “So scanning the field is not an issue. And he always tries to respond. So you do not have a reinforcement issue. Which means…”
“I have a stimulus control issue,” Sofia completes the sentence.
Sage smiles. Sofia wonders whether she can see the faintest glint of pride in Sage’s eyes. God, this woman knows how to reinforce behaviour differentially .
“And when we cannot use prompting and fading…” Sage continues.
“We create a new controlling stimulus,” adds Sofia.
“What is his strongest skill?”
“Imitation,” Sofia replies.
“Does he have functional object use?”
“Yes.”
“Then put the two together. Here is your starting arrangement.”
Sage places a cup, a flag, and a toy car on the table and invites Sofia to sit directly across from her.
“Now,” Sage says, “make the drinking gesture. Do not pick up the cup. I will be your learner.”
Sofia lifts her empty hand and makes the drinking motion. Sage immediately picks up the cup and performs the same action with the object.
Sofia knows what comes next. She rotates the objects and performs the characteristic actions for each without the object: mimicking pushing a car, waving the flag. Sage picks up each object in turn and performs the corresponding action.
Sofia can see the shape of the procedure forming. She adds the next component herself. She says “cup”, waits one second, then performs the drinking gesture.
Sage smiles. This time the glint of pride is impossible to miss.
“You got this.”
“What is next?” Sofia asks. “Do I start stimulus control transfer now, or do I build more discriminations this way first?”
“Good girl, Sofia,” Sage says, amused. “Now you are thinking like a behaviour analyst. Your choice.”
“I would build a solid set of gesture cued discriminations first, so we establish a strong reinforcement history for differentiated responses with each object.”
“Nicely done. All right, give me the next step. Imagine you now have twelve items under compound word and gesture control.”
Sofia answers without hesitation.
“New cue word, old cue gesture.
New cue: word. Delay.
If the behaviour occurs within two seconds, reinforce.
If not, deliver the old cue: gesture.”
Sage interrupts with a quiet laugh.
“Ooh, someone has a tidy intraverbal string there.”
Sofia shakes her head. God, this woman loves her verbal operants, she thinks.
When the manual has no next page
Clara scrolls through Lukas’s dashboard for the third time that morning. He has been on her review list every day lately. He is not mastering any listener target. She replays yesterday’s clip. The technician positions “ball” and “cup” correctly. AURA has recommended yet another procedure, this time a block rotation, switching target every five trials, but invariably on the switch trial, unless prompted, Lukas responds based on the last reinforced trial. She checks the intervention log. The technician has been consistent. Intervention fidelity scores are high. Response latency: one second. Reinforcers identified and effective for non listener targets. And still the same non differentiated reach.
The problem is not with these specific targets. It is with all listener responding.
AURA’s summary flashes at the bottom of the screen:
Select two new targets and initiate block rotation once more.
Next review in three sessions.
Progress status: amber to red.
Placement reassessment and funding review will be initiated if absence of progress continues.
“AURA, request supervision meeting for Lukas.”
Within a couple of hours, her supervisor appears on screen. Clara explains the issue clearly: the absence of differentiation, the repeated failures of standard protocols, the functional irrelevance of the verbal antecedent, and the looming funding review. Her supervisor listens attentively.
“Well,” she finally says, “I would keep running AURA’s protocol. The system has reviewed every evidence-based procedure available for listener discrimination. If those are not producing the skill, it may just be that Lukas cannot learn this skill.”
Clara’s stomach drops. Those words cannot learn feel like an escape hatch, another way of saying the problem lies with the student.
Her supervisor continues.
“And the funding issue is straightforward. Efficiency matters. If he’s not progressing on any listener target, we can’t really move on with the rest of the curriculum.”
“There must be another approach,” Clara insists. “There must be some arrangement we have not…”
The supervisor cuts in gently.
“Clara, AURA has evaluated all of them. Every published procedure. Sometimes this happens. Some children just do not acquire certain repertoires.”
Her supervisor does not know what to do either. She has no alternative repertoire to draw from. Without AURA, she has nothing to suggest.
But Clara knows that someone might have an answer.
“Okay,” Clara says, though she does not mean it. She takes out her personal laptop and begins drafting an email to Sofia.
Subject: Sage’s magic?
Hi ya,
Hope you are enjoying the heat and learning loads. I think of you and Sage often, but especially lately. I am struggling with a learner who is just not acquiring any listener discrimination. He is now on red alert and, unless something shifts soon, he will be moved to an alternative specialist placement. Lack of progress means his funding will be diverted to a less costly provision. Any chance you could share some of Sage’s creative strategies?
Xxx Clara
Subject: no magic wand… just behaviour analysis
Hi Clara,
It is so good to hear from you. And yes, I am learning a lot, every day. The heat is unbearable. I can see now why Sage worships the air conditioners. She looks after them as if they were pets. Yesterday I caught her saying “Good morning, my lovelies” as she climbed up the ladder and cleaned them religiously, just as she does every day.
About your learner. I really feel for you. I have been working with a little boy here who sounds very similar. No vocal behaviour and no listener discrimination when we started. He was latching on to anything except the verbal antecedent.
Long story short, I was completely stuck.
Sage helped me build something from scratch that made use of his strongest repertoire (imitation) instead of fighting his weakest. We taught him to select an item using a differentiated gesture for each object rather than going straight to giving to hand. The antecedent was arranged as word followed by gesture. Then we added a delay between the word and the gesture in a systematic way until the verbal cue began to evoke the selection and the corresponding action.
It sounds strange written like this, but in practice it is very clear.
The incredible thing is that it is working. For the first time, we have a small set of objects he selects correctly on the verbal instruction alone, without the gesture. He still requires the gesture when we introduce a new item, but the shift is real.
I am attaching the programme sheet I wrote for him. Yes, we really do write the procedures out step by step here on a word processor, and then we print them and put them in a ring binder. I hope this helps.
And please do not laugh. Sage calls it “beautifully basic”, which is her way of complimenting me and insulting me at the same time.
Kisses back,
Sofia
Finally, something different. Something new. Something that worked for a learner who sounded almost identical to Lukas. She copies the programme sheet into a clean document and uploads it to AURA’s interface.
“AURA,” she says, “generate a technician training video based on the attached procedure.”
AURA’s light pulses as it processes the request. Clara imagines the technician watching the video tomorrow, following the steps. She might even visit the centre herself, trying the procedure with Lukas. It has been so long since she has done any direct teaching.
AURA’s output appears.
Procedure review complete.
Action: denied.
Then the full message:
Clara, the submitted procedure cannot be approved.
It is not part of the published literature.
Your request to override current intervention is denied.
The “Standard Listener Programme LD5.3” sits at the top of the screen, unchanged, immutable, ready for the technician’s morning session. AURA has spoken. And within this system, AURA’s word is final.
AURA cannot approve it. Her supervisor will not approve it. And without approval, the technician cannot run it. No deviation is permitted.
She looks at Lukas’s flat graph. She knows the outcome. The funding source criteria are clear: no evidence of improvement for six months means no justification for behavioural intervention placement. And as expected, the following day, when she opens her dashboard, Lukas’s picture is no longer there. He has been transferred to an alternative specialist provision.
Reason: insufficient progress within the efficiency thresholds of the programme.
She thinks of Sofia, sitting on the floor with her little boy, her own Lukas. She thinks of Sage, guiding her through the problem, trying different arrangements until that elusive orderly relation finally emerges. It is true that AURA’s library of procedures is larger than Clara’s, or even Sage’s. But AURA cannot create anything new. It is bound by what has been programmed into its frame of reference, by the procedures that already exist.
Clara looks down at her notebook, the one Sage gave her on that first day. Only now does she understand the meaning of the line written across its first page:
Teaching means the intentional arrangement of environmental contingencies so behaviour that is not there yet can appear.
Author’s note
Thank you for reading another story from the world of AURA. I wanted to show the potential dangers of relying too heavily on a system like AURA, and the risks of training a workforce that has never learned the core repertoire of a behaviour analyst: the ability to tact contingencies moment by moment and to build stimulus control from the ground up.
We have all had learners like Lukas or Yen, children with whom we have tried every published procedure and yet nothing seems to work. This is where human generated behaviour analysis becomes essential, because only humans can create something truly new. Clara knows this now, but only because she has had a taste of what it means to problem solve when all automation is taken away.
Her supervisor, who has never experienced building behaviour from scratch, does not recognise AURA’s limitations. For her, the limitation lies with the learner. AI may make our work faster and more efficient. It may even outperform the average clinician. But productivity and understanding are not the same thing. For the experienced clinician, AI can be an enhancement, a supportive tool. If it disappeared tomorrow, someone like Sage could still teach effectively. But what about the generations who will learn entirely through AI based systems, who will never have created a procedure from the cumulation of teaching experiences? Who will have the repertoire to discern what is right, what is wrong, what will or will not work and why?
Quis custodiet ipsos custodes?
Who will watch the AI?