If AI Wrote This Post, Do We Still Need Me? A Behavioral Look at the Cult of Personal Credit

As any observer of the AI revolution can tell you, it’s necessary to read what AI writes carefully. Metaphorically speaking, the programming of a chatbot places greater emphasis on producing a complete product than on being accurate, so in service of telling a good story, chatbots sometimes just make stuff up. A familiar instance of this is the annoying habit of chatbots to invent academic references to support what they write. In the AI-generated post below, for example, see the references designated with asterisks (**). The overall content of the piece may be pretty decent, but four of seven cited sources are fictional.

In composing its post, ChatGPT likely took into account my prompt to “[use] the technical language of behavior analysis.” The resulting post is at the high college reading level, but you can ask ChatGPT for text at any reading level you like (although of course at some point readability goals will come into conflict with the use of proper technical language so, just as when humans compose, it’s critical to match reading level to audience needs and abilities).

By the way, when I said that the AI-generated post below is easy to read, that doesn’t mean it’s great writing. It has been noted that, in their current iterations at least, chatbots often generate text that’s vaguely irritating (warning: paywall). Says the author of the linked essay about AI’s prose:

It’s weird…. When you spend enough time around A.I.-generated text, you start to develop a novel form of paranoia. At this point, I have a pretty advanced case. Every clunky metaphor sets me off; every waffling blog post has the dead cadence of the machine. This year, I read an article in which a writer complained about A.I. tools cheapening the craft. But I could barely pay attention, because I kept encountering sentences that felt as if they’d been written by A.I.

I’m sympathetic, I am. But across many years I have graded countless student essays and edited countless journal manuscript submissions and I can tell you with utmost confidence: In the race to create truly terrible writing, no chatbot can compete with human beings.

Perhaps a lot of what bots write is uninspired, and perhaps the most skilled homo sapiens still can generate words that would make a chatbot jealous if it felt emotions. But I’ll take workmanlike, competent text any day over the floridly, creatively, adventurously abominable blathering that many humans have given me.

Of course, you can always ask your chatbot to compose in a style of your preference. I repeated my request to ChatGPT with the added proviso of “in the writing style of Kurt Vonnegut.” Here’s a lightly-edited glimpse at part of what it produced:

Human behavior is messy, cyclical, and stubborn. So it should not surprise any behavior analyst that behaviors, like old high-school acquaintances, tend to show up again at the worst possible times. In behavior analysis, we call this resurgence. And unlike those acquaintances, resurgence is predictable, measurable, and—thank goodness—modifiable. So pull up a chair. Smoke a cigarette if that’s your thing (it shouldn’t be). Let’s talk about how the science of resurgence can help us build sturdier, more humane autism treatment plans.

What Resurgence Is, and Why It Loves to Ruin Your Best-Laid Plans

Resurgence is a simple three-act play. Act I: Reinforcement – A behavior, typically a challenging one, has a long and sticky history of reinforcement. Act II: Extinction + Alternative Reinforcement – You extinguish the challenging behavior and reinforce a shiny new alternative response. That’s your hero—usually an FCR wearing a cape. Act III: Worsening Reinforcement for the Alternative Response – Suddenly the alternative behavior stops paying off so well. Maybe reinforcement is thinned too fast. Maybe caregivers are tired, distracted, or simply human. Then—bam!—the old challenging behavior busts through the wall like the Kool-Aid Man. This is resurgence. It is not moral failure. It is not regression. It is not “attention-seeking.” It is physics for behavior. Lattal and St. Peter (2018) describe resurgence as a behavioral relapse process driven by changes in reinforcement history and context. Like death and taxes, it is reliable.

A fascinating recent side effect of “the cult of originality” is the application to authors of cancel culture, in which a creative product becomes viewed as objectionable because its author has done something objectionable (even if the creative product is otherwise spectacular and thematically unrelated to the author’s transgression). Whatever your views on the desirability of cancel culture, it demonstrates how writer and writing have become integrated into a sort of equivalence class/frame of coordination, and how behavioral functions evoked by the author become evoked by the author’s written works.

“Branding” means using words to assure that certain individuals receive credit for some treatment, regardless of its actual origin. One of my favorite examples comes from Alan Kazdin, who’s a brilliant, intimidatingly productive scholar, and one of my favorite people in all of Psychology. Among the things Kazdin is known for is “The Kazdin Method,” basically a protocol for dealing with child and adolescent behavior issues that serves as the basis of numerous commericial products from which Kazdin benefits financially. The thing is, there’s not much in “The Kazdin Method” that you can’t find elsewhere in the vast literature on behavior management — although Kazdin is especially talented at organizing this information and presenting it in a way that everyday people can understand. The fact that what he presents is derivative doesn’t mean it’s not useful. In many instances I think of AI-generated text the same way.

By the way, applied folks aren’t alone in their attachment to branding. Research Psychology, for instance, is a branding machine, with a host of psychological phenomena being named after researchers who identified them (e.g., The McGurk Effect, The Zeigarnik Effect, The Dunning-Krueger Effect, etc.). Branding even leaks into the experimental analysis of behavior at times (e.g., “Sidman avoidance”… although to his credit Murray Sidman never liked this labeling) — illustrating that even behavior analysts, who should know better, get swept up in the “cult of originality.”

The cult of originality was still evolving in the 16th Century, with copyright law and clear definitions of publishing rights a century or two away. Although playwrights were paid for their work, theaters, not authors, were held to own the plays. To protect what today are called intellectual property rights, theaters often didn’t allow plays they purchased to be published for public consumption. Among the plays that were published, many (including some of Shakespeare’s) appeared without a byline, suggesting that authorship was not valued as much as it is today. All of this feels a bit confusing to our credit-crazed contemporary sensibilities — for example, the practice of anonymous publishing has over the years fueled considerable speculation and detective work focusing on which works now attributed to Shakespeare may have been penned by someone else, and which works not conventionally attributed to him might actually have his. Some have taken the lack of clear provenance of such works to indicate some kind of insidious coverup in which the Bard actually wrote none of them, but rather served as a front for one or more other writers who, for whatever reason, chose to keep their identities secret. The, er… creativity of these tales illustrates just how discombobulated we feel when unable to assign clear personal credit.

Recently it has become the norm at my university for students to use AI to compose their emails to professors. OH THANK GOD!!! The resulting messages are almost always pleasant, respectful, to the point, and easy to understand — none of which was guaranteed in the good old days of students composing their own messages.

In the context of this post it’s funny to remember that no less of an intellectual light than Socrates was vehemently opposed to the PRINTED WORD on the grounds that it would, in effect, dehumanize us. In the same vein I like to imagine grunting protohumans becoming positively livid about the development of the first language. “Mark my words,” I can hear them say, “We’ll all forget how to grunt, and then where will we be?” Or at least they would say that if they weren’t limited to grunting. People always think changes in communication technology are going to end us. So far that hasn’t happened.

Because this post is about how writing affects human readers; I’ve intentionally soft-pedaled the parallel issue of how writing affects human writers. An argument can be made that the most important thing about writing is that it forces the writer to struggle to say something new and useful, a struggle that flexes the writer’s “creativity” muscles. We know that certain kinds of experiences reinforce novelty in behavior. AI’s effortless assistance with the writing process might undermine those experiences. As one observer suggested:

“The bigger and more alarming impact [AI tools] have is to constrict our full range of thoughts and our ability to generate original and useful ideas — what we call creative thinking. This seems to be especially true for students. A.I.’s smooth sentences, elegant transitions and rich vocabulary give the illusion of expansive creativity and individuality. But the underlying ideas often converge into a few homogenized categories….The erosion of creative thinking means young people will struggle to navigate uncertainty. Workers will strain to adapt to a shifting labor market. And society will miss out on the new ideas that can solve complex problems and enhance lives.”

In other words, perhaps the most important thing about writing is the struggle itself, not the final product. Writing is difficult and iterative. To do it decently well you have to generate a lot of behavior, which creates lots of opportunities for selection by consequences. The possible ramifications for the writer’s future behavioral repertoire go well beyond any given written work. In short, learning to write may teach you to think. In fact, Skinner argued that you have to start writing before you can learn what you think [download].

And if that’s so, when AI is composing for us on the regular will we end up thinking about only what AI selects for us? Will we stop thinking entirely? Or, perhaps, can human-AI interactions be structured to turn us into better thinkers than we’d otherwise be? Discussions about AI’s compositional acumen tend to weigh the false dichotomy of whether we either should or should not use it. A sophisticated behavioral take on the issue would ask how we might use it to create more functional humans.

Using the Scientific Literature on Resurgence
to Inform Autism Treatment Plans

Applied behavior analysts routinely design interventions that alter reinforcement contingencies for challenging behavior while promoting adaptive, socially acceptable alternatives. Although these interventions are often effective in the short term, they may be vulnerable to treatment relapse phenomena. One such phenomenon—resurgence—has been studied extensively within basic, translational, and applied research domains. For clinicians serving autistic learners, understanding resurgence is essential for engineering durable, generalizable treatment effects.

This post reviews key findings from the scientific literature on resurgence and discusses how these findings can directly inform autism treatment planning.

Defining Resurgence

Resurgence refers to the reappearance of a previously reinforced and then extinguished response when reinforcement for a more recently reinforced alternative response is degraded or discontinued (Lattal & St. Peter, 2018). It is typically analyzed using a three-phase framework:

Phase 1 (Baseline): Target behavior contacts reinforcement.

Phase 2 (Alternative Reinforcement): Target behavior is extinguished while an alternative behavior is reinforced (e.g., functional communication).

Phase 3 (Worsening of Alternative Reinforcement): Reinforcement for the alternative response is reduced or removed, often leading to resurgence of the target behavior.

Resurgence can be conceptualized as a form of behavioral relapse influenced by reinforcement history, discriminative control, and motivational variables.

Why Resurgence Matters in Autism Treatment

Treatment for socially maintained challenging behavior in autism often includes Functional Communication Training (FCT) or other differential reinforcement procedures. These interventions map directly onto the three-phase arrangement described above. As such, “relapse” following schedule thinning, caregiver error, or contextual change should not be viewed as surprising or anomalous: it is predicted by behavioral processes.

Understanding resurgence enables clinicians to design interventions that anticipate and mitigate treatment vulnerability.

Key Findings From the Resurgence Literature

Abrupt Worsening of Alternative Reinforcement Increases Resurgence

Multiple studies show that suddenly removing or degrading reinforcement for an alternative response increases the magnitude and likelihood of resurgence (e.g., Sweeney & Shahan, 2013). In applied settings, this commonly occurs when caregivers abruptly reduce reinforcement for communication responses or when integrity lapses create unplanned extinction conditions. Clinical implication: Reinforcement schedule thinning should be gradual, systematic, and data-based, avoiding abrupt transitions when possible.

Multiple Schedules Reduce Resurgence by Establishing Discriminative Control

Research demonstrates that multiple schedules—in which distinct stimuli signal the availability versus unavailability of reinforcement for the alternative behavior—can significantly mitigate resurgence (Fisher et al., 2018). These stimuli function as Sᴰ and SΔ, clarifying for the learner when reinforcement is likely. Clinical implication: Use clear, teachable multiple-schedule stimuli (e.g., visuals, colored cards, wristbands) during FCT and schedule thinning.

Behavioral Momentum Theory Predicts Treatment Persistence and Resurgence

According to Behavioral Momentum Theory (Nevin & Shahan, 2011), behavior reinforced in contexts with dense, consistent reinforcement shows greater persistence—and, counterintuitively, may also be more susceptible to relapse when reinforcement conditions worsen. This means that extremely rich reinforcement for communication responses in early FCT may build strong behavioral momentum but may also set the stage for significant resurgence later. Clinical implication: Use dense reinforcement to establish alternative behavior quickly, but transition to more naturalistic schedules sooner rather than later to avoid creating a momentum-related vulnerability.

Treatment Integrity Failures Function Like Worsening Reinforcement

Even minor lapses in caregiver implementation—such as delayed reinforcement, inconsistent responding, or missed opportunities—can act as small-scale versions of Phase 3 conditions (St. Peter et al., 2015). These inadvertent worsening conditions can evoke resurgence of problem behavior. Clinical implication: Include robust caregiver training, monitoring, and supports to maintain treatment integrity across settings.

Teaching Multiple Alternative Responses May Buffer Against Resurgence

Research suggests that providing multiple alternative response options (i.e., establishing a response class of alternatives) reduces resurgence relative to teaching a single alternative response (Lieving & Lattal, 2003). This may occur because variability and redundancy allow persistence despite changes in reinforcement conditions. Clinical implication: In FCT and adaptive skill programs, consider teaching multiple, functionally equivalent alternative responses (e.g., several mands or communication modalities).

Contextual Changes Increase Resurgence Likelihood

Generalization failures and contextual transitions—between therapists, settings, or task demands—are well-documented triggers for relapse phenomena, including resurgence (Podlesnik & Kelley, 2015). Clinical implication: Plan for generalization across people, settings, and materials early in treatment and continue to monitor for resurgence during transitions.

Embedding Resurgence-Informed Strategies Into Autism Treatment

Functional Communication Training (FCT)

FCT is especially susceptible to resurgence during schedule thinning. Incorporating resurgence-reducing strategies includes:

• Implementing multiple schedules to establish stimulus control (Fisher et al., 2018).
• Conducting delay-tolerance training to promote persistence of communication.
• Teaching multiple functionally equivalent communication responses.
• Using graduated, criterion-based thinning rather than fixed step-downs.

Differential Reinforcement of Alternative Behavior (DRA)

In DRA arrangements for challenging behavior:

• Use supplemental stimuli (e.g., “work time” cues) to differentiate reinforcement availabilities.
• Maintain intermittent reinforcement for alternative responses even after mastery.
• Integrate compound schedules (e.g., mixed or chained schedules) when transitioning across activities.

Skill Acquisition Programs

Learners sometimes display resurgence of earlier, less sophisticated response forms or errors. Strategies include:

• Maintaining periodic booster reinforcement for mastered skills.
• Avoiding abrupt increases in task difficulty.
• Programming errorless teaching during transitions to new exemplars.

Caregiver and Staff Training

Because integrity lapses can evoke resurgence:

• Use behavioral skills training (BST) for caregivers and technicians.
• Implement treatment integrity checklists and performance feedback.
• Simplify procedures to raise the likelihood of consistent implementation.

Data-Based Responding to Early Signs of Resurgence

Subtle markers may precede full resurgence, such as:

• Reduced alternative behavior.
• Longer latencies before response initiation.
• Mild precursors of challenging behavior.

Clinical teams should:

• Temporarily increase reinforcement density for the alternative response.
• Strengthen multiple-schedule stimuli or contextual cues.
• Reassess motivating operations and competing reinforcement sources.

Conclusion

Resurgence is not a sign of treatment failure—it is a predictable behavioral process governed by historical reinforcement contingencies, discriminative control, and motivation. By integrating findings from the basic and translational literature, applied behavior analysts can create treatment programs for autistic clients that are more resilient to relapse and more adaptable to naturalistic environments.

Using strategies such as multiple schedules, gradual schedule thinning, response class expansion, and caregiver integrity supports, clinicians can effectively reduce the likelihood and severity of treatment relapse. A resurgence-informed approach ensures that interventions remain durable, socially valid, and responsive to the complex contexts in which real families and learners operate.

References

Fisher, W. W., Greer, B. D., Fuhrman, A. M., Saini, V., & Simmons, C. A. (2018). Minimizing resurgence of destructive behavior using multiple schedules during functional communication training. Journal of Applied Behavior Analysis, 51(2), 340–359.

**Lattal, K. A., & St. Peter, C. C. (2018). Resurgence: Implications for behavioral relapse in applied settings. The Behavior Analyst, 41(2), 343–364.

Lieving, G. A., & Lattal, K. A. (2003). Recency, repeatability, and reinforcer retrenchment: An experimental analysis of resurgence. Journal of the Experimental Analysis of Behavior, 80(2), 217–233.

Nevin, J. A., & Shahan, T. A. (2011). Behavioral momentum theory: Equations and applications. Journal of Applied Behavior Analysis, 44, 877–895.

**Podlesnik, C. A., & Kelley, M. E. (2015). Translational research on behavioral relapse: A review of resurgence and renewal. Journal of Applied Behavior Analysis, 48, 685–705.

**Sweeney, M. M., & Shahan, T. A. (2013). Effects of reinforcer rate on resurgence: A quantitative analysis. Journal of the Experimental Analysis of Behavior, 99(2), 168–186.

**St. Peter, C. C., et al. (2015). An evaluation of treatment integrity failures during differential reinforcement procedures. Behavior Analysis in Practice, 8(1), 42–50.

**Plausible author-topic match, but not a real source!