Copilots and the Future of Applied Behavior Analysis Therapy for Autism

4 min read

A futuristic therapy room with holograms and toys.

Introduction

Autism treatment remains a complex and debated topic, yet Applied Behavior Analysis (ABA) stands out as the scientifically backed gold standard for reducing harmful symptoms and teaching essential skills. Despite its effectiveness, the demand far exceeds the supply, leaving hundreds of thousands of children on waitlists across the United States.  

The grey regions on this map indicate areas where children with autism are present but lack access to Board Certified Behavior Analysts (BCBAs) to provide treatment. 

Enter the era of Artificial Intelligence (AI), which is reshaping various industries, including Applied Behavior Analysis. As ABA and AI integration occurs, Board Certified Behavior Analysts (BCBAs) will enjoy highly capable, skill-augmenting, AI co-pilots just like programmers and investment bankers already use. Even teachers and schools are jumping on board!

But how will this tech assist BCBAs, who do such highly technical and complex work? Let’s explore that possibility!

AI is not just automating tasks; it is enhancing or surpassing human capabilities (World Economic Forum)​. The following graph illustrates how AI exceeds human baseline performance in various domains, given thorough training data. Astute observers will also note that the time for AI to match human performance is also decreasing on a nearly exponential scale.

  Human baseline performance is becoming reachable by AI-counterparts in 1–2 years now, given thorough enough training data. https://aiindex.stanford.edu/report/

We can reason straightforwardly, then: if AI can perform as well as humans on verbal reasoning, visual processing, and math — they will soon outperform human BCBAs on tasks requiring those skills, and complete them in a fraction of the time that it takes a human.

What tasks do BCBAs do almost everyday that fall within those domains?

  • Writing treatment plans, behavior plans, and payer reports
  • Creating graphs and performing analyses
  • Interpreting graphs and data

Limitations and Solutions

You may be thinking, “hold on, GPT-4 or Opus can’t do that yet,” and you’re right. Current large language models (LLMs) certainly have some challenges — they can hallucinate false information, their context windows are too small for ABA datasets (i.e. can’t ‘think’ about all necessary info at once), and they don’t understand Applied Behavior Analysis because their training datasets are too broad and uncalibrated.

However, it is important to realize that these issues are symptoms of the new technology’s growing pains, rather than insurmountable obstacles. Computer scientists have already devised a way to store and retrieve client-specific and ABA-domain specific data, eliminating hallucinations. The context window may be small, but no matter — dozens of separate instances (“agents”) of a language model are able to all collaborate on a large task (see https://www.crewai.com/) to provide adequate redundancy and speed. Last but not least, models are able to be fine-tuned on ABA-based training data to perform extremely well on ABA-related tasks.

As you can see all of these technologies are not only possible, but solutions are already developed!

So, will this replace BCBAs? Not without these AI also being able to collect their own data, make their own decisions, and automatically incorporate feedback. However, in the mean time, it will absolutely be responsible for creating a sea change in the workflow for BCBAs.

AI-Augmented BCBAs: Enhancing Efficiency and Reducing Burnout

Just glancing at Reddit’s Behavior Analysis threads over the last couple of years reveal some common themes:

2 hours per day for graphing, reporting, and revising goals/targets
10 hours per initial assessment report/treatment planning
New BCBAs need help: 2–3 hrs. of work needed to improve their reports

If the theme isn’t obvious, BCBAs are largely burnt out (70%+!). The never-ending and repetitive report writing, remembering all of the separate payer rules, and RBT turnover are largely responsible. Not to mention the sheer amount of data that can be analyzed is larger than any human could consider, truthfully, and for some this is also source of overwhelm or analysis paralysis.

Let’s critically evaluate a potential solution by examining the pros and cons of BCBAs (or BCaBAs) utilizing AI Copilots for repetitive tasks::

Pros:

Cons:

  • Additional costs incurred? (BCBAs will replace lost writing/analysis “office BCBA’ hours with other necessary billable activities or with additional clients)
  • BCBA loss of skills? (Some old skills, sure. Along with the rest of the workforce that adopts AI. What’s important is prioritizing the correct skills to keep and upgrade)
  • Privacy and regulatory concerns? (All tools developed and released for BCBA-use must of course be HIPAA compliant and abide by FDA guidelines)

Looking ahead, what additional features could AI offer to BCBAs?

New insights into behavior?

Yes — take a look at what a machine learning model recently found was the best predictor (highest “Feature Value”) for the necessary intensity for ABA therapy:

  Bathing ability is more predictive than age and amount of prior ABA when predicting ABA treatment requirements. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9981822/

This is an isolated example, but the point stands — this predictive ‘feature’ existed in that dataset, and humans would possibly never find it on their own. I’m betting many analogous examples are present and waiting to be uncovered!

Managing the Great AI Integration

If this sounds like a lot to take in, don’t worry — it’s only the beginning ! If you’re reading this in 2024, you’re still early — large scale investment and development has only begun!

Generative AI investment has only just begun, rising from $2.8 Bn in 2022 to $25.2 Bn in 2023.

The OECD and other organizations stress that highly skilled jobs are at the highest risk for AI-integration related disruption, hence the importance of preparing for and adapting to AI-induced changes.

Due to the extraordinarily rapid pace of this change, experts in the Harvard Business Review are recommending that employers begin to budget for AI upskilling — boldly proclaiming it as the “responsibility of every leader and manager.”

For BCBAs and similar professionals, this means adapting to use AI as a tool that complements their expertise, allowing them to successfully manage larger caseloads, create better outcomes, and/or focus on other aspects of their work that significantly impact those who we care about the most — the clients.

What do you think? Leave a comment about it, let’s shape the future of AI and ABA to be the brightest one possible!

Josh Farrow With over 15 years of expertise in behavior analysis, I specialize in optimizing therapeutic practices and integrating technology to enhance service delivery. I'm dedicated to pushing the boundaries of behavioral science through innovative technology. Let's connect and explore how we can collaborate to make a meaningful impact in the ABA community! https://calendar.app.google/Dpw9icf17xhfCmrg6

Leave a Reply

Your email address will not be published. Required fields are marked *