Chris Sullens is the Chief Executive Officer of CentralReach and is leading the company in its mission to provide an end-to-end platform of software and services that enable Applied Behavior Analysis (ABA) clinicians and educators to produce superior client outcomes for people with autism and related disorders.
The company has just unveiled its new generative AI assistant, cariTM. The leading team of clinicians and data scientists at CentralReach believes that cari will be the most powerful and secure AI-assistant for providers serving the autism and intellectual and developmental disabilities (IDD) space
Could you share some insights as to how CentralReach is currently serving individuals with autism and IDD to deliver evidence-based and personalized care, at scale?
CentralReach provides the leading software and services platform to help children and adults diagnosed with autism and related intellectual and developmental disabilities (IDD) – and those who serve them – unlock potential, achieve better outcomes, and live more independent lives.
Our software automates the end-to-end care delivery process that occurs both by therapists in a clinical setting as well as by special education staff who are working with children and adults in the school setting. Through that automation, our customers have been able to expand services, open new clinics, serve more students and thanks to that, CentralReach’s suite of products has helped over 150,000 users serve nearly 500,000 children and adults on the autism spectrum.
In what ways has CentralReach historically leveraged machine learning and AI?
In November 2022, we launched an AI-powered scheduling solution that helps our customers automatically optimize calendars using a number of variables to make room for new clients. Like so many sectors of healthcare, scheduling can often be a full-time job – but particularly in ABA care delivery, scheduling services which can often be as high as 40 hours per patient per week, is time-consuming. CR ScheduleAI, which is powered by our cari™, CentralReach’s AI assistant, allows schedulers to easily build and maintain complex provider schedules in a secure, HIPAA-compliant way based on client and staff availability, distance, key treatment criteria, and provider skill sets. Since the product launched last year, we estimate 20%+ more autistic children and adults are receiving appointments.
On August 8th, 2023, CentralReach announced a generative AI Assistant designed for autism and IDD care: cari. Can you share with us how this AI Assistant will be used for AI-assisted patient summaries?
Absolutely. This use case is something we are very excited for. With a few quick inputs from the clinician, cari will provide a well-written session summary for clinical staff to review and edit before submitting. The solution will transform a task that currently takes clinicians hours to complete to one that requires only time to review, check and submit, freeing up substantial time for clinicians and greatly reducing the delay between session completion and payment receipt.
What are some other ways that cari will be used?
There are a myriad of ways that cari can be used from automating internal processes and making it easier for customers to get help to helping decipher which protocols lead to the best outcomes. But the two near-term applications of cari beyond the session summaries and scheduling previously mentioned include 24/7 clinical test prep support and fast-track assessment recommendations.
For test prep support, cari will help aspiring registered behavior technicians (RBTs) prepare for the BACB exam. A key pain point for test takers and those that are mentoring and supporting them is the myriad of questions that surface as prospective RBTs work their way through the 40-hour course. By training cari with CR Institute’s market leading test prep tool and directing the AI assistant’s knowledge pool to CentralReach’s nuanced and comprehensive database, prospective RBTs can ask cari questions they would otherwise need to direct to supervisors or be hesitant to ask.
When it comes to autism-specific assessments such as the ABLLS- R® and AFLS®, the challenge for every clinician is quickly determining the best place to start. By providing a few key inputs about the learner, cari will quickly analyze over 2,400 skills covered in the ABLLS-R and AFLS protocols and recommend the optimal starting point and a personalized plan for the learner in minutes for the clinician to review.
Can you discuss the size, data points, and quality of the data that was used to train the cari AI?
We have well over 3 billion data points in our proprietary dataset – a number that is growing by 1 billion data points on an annualized basis. Within our initial use cases, we have focused and trained cari on a subset of that dataset. And, it’s this proprietary dataset that stands out from applications applying generic large language models (“LLMs”) or startups utilizing subscale or aggregated data models that don’t include nuanced information only applicable to the autism and IDD care industry. So, given the volume of our dataset, which includes financial, operational, and clinical data points from organizations of every size, and its dedicated, secure and HIPAA compliant LLM environment, it truly is the only dataset of its kind that reflects the autism and IDD care industry.
With hallucinations being one of the most prolific problems with Generative AI, how is CentralReach tackling the challenge of minimizing or eliminating hallucinations?
Oh it’s been fun. While hallucinations are very difficult to fully eliminate, we’ve been able to greatly reduce the potential for them by focusing cari’s training on our proprietary data set and introducing guards on what the model can use, which severely limits generic data that’s housed in other LLMs. On top of that, we’ve been able to tune model parameters to favor precision over free-form creativity.
Additionally, we combine generative AI with robust human-in-the-loop checks and quality control processes. We have an extensive team of autism and IDD care experts called Board Certified Behavior Analysts (BCBAs) who review all outputs to catch and correct any potential inaccuracies or false information before inclusion in our applications.
We are also continuing to explore techniques like human-AI collaboration, where generative models make draft suggestions while people provide feedback and oversight to further refine the output. By emphasizing supervision and context-specific training, we aim to harness the upsides of generative AI while instituting guardrails against hallucinations.
Lastly, we are committed to delivering reliable, ethical software solutions to our users and will only incorporate new technologies once we are confident they can meaningfully improve experiences without compromising quality or trust.
For readers concerned about AI replacing health practitioners, can you discuss how cari is designed to work alongside and augment human workers, instead of replacing them?
It’s a great question. The reality is GenAI cannot replace what autism and IDD care professionals do but it can support them. The goal is to greatly reduce the time it takes for them to perform necessary back office and clinical tasks (that in many cases are not reimbursable) and provide decision support on many day-to-day tasks to help them improve outcomes for their clients and patients.
Is there anything else you want to share about CentralReach or cari?
What we’ve shared today is just the tip of the iceberg. There are millions of children and adults on the spectrum and many of those are waiting for services as there continues to be a shortage in clinicians available to deliver care. And, we believe software and innovation can be one of the keys to solve this problem. So, we are on a mission to continuously improve our software and deliver new innovation that enables therapy providers and schools to increase access to care and improve outcomes for the autism and IDD population. Stay tuned on the latest improvements at www.centralreach.com.
The post Chris Sullens, CEO of CentralReach – Interview Series appeared first on Unite.AI.