Applied Behavioral Musicology: Classic Behavior Analysis Publications, In Song

Derek D. Reed, Institute for Behavior Resources

Tom Critchfield, Illinois State University

#4 in the AI Diaries series. Previous Posts: 1 2 3

This post is a companion of sorts to an April 1 offering in which we used artificial intelligence tools to imagine a lost recording of historical significance. Here we delve deeper into the capabilities of the same technology.

---

I Wonder…

You could, if you wish, think of the present post as an existential inquiry into how artificial intelligence (AI) is encroaching upon domains (like musical creativity) that were once uniquely for human beings.

Or, you could view this post as extending an argument, advanced in previous “Something Interesting” posts, about harnessing media other than books and journals to disseminate behavior analysis.

OR, third option, you could not think so much and just read on to see what happens. We recommend the mindful approach, but you do you.

---

I wonder, wonder who, mm-badoo-ooh, who,
Who wrote the Book Of Love
– from “Book of Love” (1958) by The Monotones

“See what happens” was our primary motivation for exploring how autonomous AI agents would handle the challenge of generating pop-music-style songs based on memorable contributions to the behavior analysis literature. We’d tinkered with the relevant tech a bit for our last post, and being suckers for approaching behavior analysis in unconventional ways, we couldn’t resist taking things further.

Going into this project, we were of two minds. On the one hand, we thought it would be foolish to bet against today’s AI, which has impressive capabilities and already is invading popular music charts with compositions that most listeners cannot distinguish from human-made. On the other hand, we recognized that our challenge was one that might push even the craftiest human composer to the brink, because, well, let’s face it, there’s nothing catchy or lyrical about how behavior analysts normally communicate about their science (also see Note 1).

NOTE 1: THE LOW BAR SET BY HUMAN LYSICISTS (click to expand)

Maybe it’s the subject matter we gave ChatGPT to work with, but sometimes we had difficulty steering it away from erudite-but-clunky wording. It has a knack for producing words and phrases that convey an abstract air of profundity without clearly communicating anything. To be completely fair here, carbon-based life forms don’t always create spectacular poetry either. To illustrate, here’s a little ditty we threw together in the Suno app that, depressingly enough for the human race, employs lyrics from songs composed by human writers and made popular by human listeners (and trust us when we say that it took almost no time to compile the raw material). We think that AI’s lyrical efforts will look pretty shiny by comparison.

But our purpose was not to mount a John Henry-esque person-versus-machine competition, so in true inductive fashion we will simply tell you what we did and show you what AI gave us in response.

Methods: Creating The Tracks

We based our human-computer compositional collab on a variety of published behavior analysis sources: familiar classics, underappreciated gems that tell an important story, and a couple of other things from the behavior analysis literature. The songs you’ll hear below were created with limited human input beyond feeding those sources into AI tools. The two of us independently produced several songs using this method (see Note 2):

1. We fed the source publication into ChatGPT and asked for lyrics based on it.
2. We fed the lyrics into the music-generating app Suno and asked for a song.

NOTE 2: MORE TECHNICAL INFO (click to expand)

• About musical styles: ChatGPT will offer to create lyrics that are compatible with a music genre of your choice. We occasionally accepted the offer but usually not. In Suno, we usually but not always specified a musical style with which to pair the lyrics, aiming for variety across songs. However, given our wordy academic source material we steered away from genres that are light on verbiage, and we make no claim to have employed a representative variety of genres.

• Our original intent was to showcase purely digital creativity, but we don’t wish to oversell. Sometimes, as hoped, AI delivered what we regarded a decent track right away. Other times we had to negotiate with the tech a bit, and there is in fact an art to asking a bot for exactly what you want. Long story short, on occasion we tweaked uninspired ChatGPT lyrics or asked Suno for multiple versions of a song when the first ones turned out forgettable. Even though some tracks benefitted(?) from a human touch, AI absolutely did the compositional heavy lifting. For most tracks, the time we spent tinkering was less than the time spent waiting for bots to process our requests (response times could range up to about 5 minutes per request). And for authenticity’s sake we did not try to fix everything that we thought could be improved. In particular, Suno is prone to verbal hiccups, so sometimes in the tracks below you can catch common words being oddly pronounced (e.g., Suno really struggles with “analysis,” which can be a problem in songs about behavior analysis). These instances stand as testament to the fact that AI cannot yet always fully replace a human. Not yet.

• If you’re a music person you’ll probably regard “our” compositions as pretty conventional, meaning no matter a track’s genre, you’ve probably heard something a lot like it before. This is a standard gripe about AI-authored work, and that’s not too surprising given that autonomous AI agents are trained up by feeding them with mountains of existing examplars. Just as with living organisms who learn under relatively stable contingencies, the resulting “behavior” can reflect a degree of stereotypy (meaning songs can sound derivative rather than groundbreaking). Special learning experiences are needed to spawn real novelty. Also, full disclosure: To some observers, AI “creators” that are trained on and can imitate human creations raise concerns about intellectual property rights and more.

Results: The Definitive Behavioral Musicology® Playlist!

The songs presented here vary considerably in their approach to the subject matter. Some are more “preachy” (conveying a general message rather non-technically) while others are more “teachy” (explaining some of the fundamentals of behavior analysis). That’s just how it turned out. Beyond that, sure, we have thoughts: about how reliably the tech works, about the lexical and musical quality of what it creates, and so forth. And, if pressed, we would acknowledge that a couple of compositions may have spun in… unexpected directions.

But our overarching reaction is to be stoked about advancing the budding young field of behavioral musicology® by presenting you with nearly two hours of the best behavioral beats you can bop to this side of the BACB.

These songs prove that it’s possible — actually very simple with AI technology — to create enjoyable alternatives to abstruse academic articles and tedious old textbooks (for perspective on this point, see Notes 3 and 4).

The playlist can’t yet replace a respectable graduate education, of course, but we’re pretty sure it’s already at least as pedagogically valuable as a Registered Behavior Technician® training course (and it’s free, and clocks in at 38 hours shorter).

NOTE 3: A CATCHY ATTENTION GETTER (click to expand)

As an exercise in applied behavioral musicology®, this post overlooks an important preliminary step. Before anyone would bother to seek out information, musical or otherwise, about behavior analysis they would first require reinforcer sampling — that is, an invigorating introduction to what it is and how exciting its application can be. In theory, music can convey that in a straightforward, enjoyable, and easy to digest way — say, something even your grandmother would understand and tap her toe to. Example: An earlier post described how one elementary school student quickly formed lifelong memories of the basic structures of a cell by listening to a three-minute song called “Cell City.”

How to create something like that for behavior analysis? As we’ve demonstrated here, crafting captivating tuneage is no great challenge with AI assistance. All we need, then, is an invigorating introduction to set to music… And guess what? Since the 1970s our professional organizations have led the way by carefully considering how to present behavior analysis to the public. Problem solved, right? With a little appropriation, it’s easy to envision our discipline’s answer to “Cell City.” And so, without further ado and with lyrics from by a certain high-profile organization, here’s a catchy example we threw together to illustrate the application of behavioral musicology to the marketing of behavior analysis itself.

NOTE 4: DATA DON’T LIE (click to expand)

We might be joking, just a little, about using song as a means of disseminating behavior analysis, but we’re serious in suggesting there’s value in breaking free of media that house our scholarly communications. A recent “Something Interesting” post showed that a lot of what’s written about behavior analysis scores unfavorably in a sentiment analysis (i.e., it’s probably unpleasant to lay folks). Below is an example, showing the sentiment scores (black triangles) of 10 published articles that advocated for more pleasant communication about behavior analysis. Ironically, many of these score as unpleasant in a sentiment analysis. Superimposed are the sentiment scores (yellow circles) of AI-generated lyrics for a randomly-chosen 15 of the songs presented here. Yes, there are a few sourpusses in the mix, but for the most part the lyrics are sunnier than verbiage produced by people who expressly set out to spread good cheer. This isn’t entirely unexpected since AI is trained on massive samples of human output, and when humans who are not behavior analysts output language it tends to skew emotionally positive.

Yeah, sure, we know you’d prefer to talk about the threat AI poses to human creators, but we found our digitally-collaborative creative experience to be not diminishing but rather empowering. Of course, we are musical morons who’ve never written an actual song in our lives, so we have nothing to lose when the machines take over. We’ll probably feel differently when AI replaces behavior analysts, which can’t be far in the future. Regardless of whether AI supplants or enhances human expertise, however, this frames a rich topic for behavior analysts: After all, what makes bots interesting includes that humans interact with them, so behavioral issues always are involved. We hope this post gets you thinking about some of those.

But for now, take a beat, ditch the Deep Thoughts, and get your bow-chicka-wow-wow on with these inspirational ear worms. Go ahead, download ’em for personal use (just click the triple dots to the right of each track). Then belt ’em out on your commute. Teach with ’em (see Note 5). Play ’em in your clinic waiting room. Use ’em for mood music at your poster or ABAI Expo booth. Maybe choose a track as theme song for that new Behavioral Musicology® Special Interest Group you’ve been inspired to start. Oh, and ABAI, you now have a perfect soundtrack for the extended slide show that always precedes the Presidential Address at the annual convention.

In short, the possibilities are endless.

NOTE 5: PEDAGOGY (click to expand)

To support that college course you’re going to organize around the playlist, we’ve divided it into four handy thematic units below. A valuable conceptual exercise — especially these days, when the classics aren’t taught much — would be for students to carefully evaluate how accurately each track reflects its source material. Imagine a verse-by-verse critical dissection annotated with specifics of the source material. That’s an essay per track, and we guarantee a student who gives this “curriculum” a serious effort will come away with lasting memories of those classic sources. Not to mention, perhaps, a better sense of how to accessibly communicate about technical material.

In the poll following the playlist, tell us your favorite track. In the Comments box at bottom, tell us what you think about the general topic of AI-assisted creativity, or about these tunes specifically. We might just award a prize for the best comment. And if you’ve created your own track (through whatever means) that you think conveys something important about behavior analysis, provide a link in the Comments box so readers can enjoy your composition.

Hear this playlist in HD uninterrupted in Soundcloud, along with, at absolutely no extra charge, several bonus tracks (see the Liner Notes below for a list). But warning: At the end of a playlist Soundcloud annoyingly feeds you tracks from wherever the hell it feels like. Sorry about that.

UNIT 1: Laudable Lab Work

UNIT 2: Conceptual Foundations

UNIT 3: Iconic Applications

UNIT 4: Reflections on the Health & Well-Being of Behavior Analysis

LINER NOTES: List of Bonus Tracks in Soundcloud

MASHUP: “Farewell my LOVELY! (alternate version)” + “I Love Towels”

GONE GONE GONE Based on Azrin, N., & Foxx, R. M. (2019). Toilet training in less than a day. Gallery Books.

ODE TO THE OLDEST JABA PAPER TO HAVE NEVER BEEN CITED ACCORDING TO WEB OF SCIENCE Based on Fargo, G. A., & Behrns, C. (1969). Rapid computation and pupil self-recording of performance data. Journal of Applied Behavior Analysis, 2(4), 264. PDF [Please help to rectify an historical travesty by citing this paper as often as you are able!!!!]

WHY WE ARE NOT ACTING TO SAVE THE WORLD (death metal version for realists)

MASHUP: Who We’re Meant to Be + Why We Are Not Acting To Save the World (death metal version)

---