Creating the ‘Always Learning’ Virtual Patient
Until recently, purchasing healthcare simulation equipment meant that clinical simulation educators were buying the simulator as is at the very moment of purchase. Any design quirks or lack of functionality would simply remain with the buyer and their simulator throughout the simulator’s lifespan. Over time the simulation educator would uncover more things they wished their simulator could do, or maybe potential changes in curriculum meant their now old simulator companion simply wasn’t up to the task. This generally sparked the time to start seeking out a replacement and the cycle would inevitably repeat itself. Such is the “hardware” way.
A Constantly Updating World
With the release of the iPhone in 2007, consumers began to be conditioned to expect a continually improving product with annual software updates that would add new functionality and address annoying bugs. No longer was the phone that you purchased on day one, the same phone two years later from a software perspective.
This behavior extended beyond mobile phones to TVs, refrigerators, cars – and yes – even human patient simulators. No longer is both the software and hardware “static” on these human replicas. Instead, the software is constantly evolving to better meet the needs of the simulationist, without necessarily requiring a new simulator purchase.
Behind the scenes, teams of engineers work to develop these new functionalities that are released in one, maybe two, annual software updates. While a far cry from the past, what if the software could teach itself to improve on a monthly, daily, or even hourly basis?
Independent Machine Learning
You’ve probably heard of the term “machine learning” thrown around as the tech world continues to expand on artificial intelligence capabilities. Expert.ai defines machine learning as, “An application of AI that enables systems to learn and improve from experience without being explicitly programmed. Machine learning focuses on developing computer programs that can access data and use it to learn for themselves.”
This is exactly what the team at PCS set out to do with their conversational virtual patient platform, PCS Spark. Launched back in 2019 (before anyone had heard the word COVID), Spark was a VR-based virtual patient that, for the first time, allowed learners to interact verbally with a virtual patient, just as they would with a standardized patient or an actual patient.
Those were the early days for the PCS Conversation Engine powering Spark as the user base was just beginning to grow. The initial questions were used as a dataset for the AI-training model and were mainly composed of an internal team of former standardized patient educators. Think of this as “supervised learning,” with the team setting the initial boundaries of what the conversation engine should be able to listen for and respond to.
Generally speaking, Spark was able to understand questions related to the history of present illness, past medical history, and some questions relating to social history and associated risk factors. This was enough for some innovative instructors to see the promise and begin adopting this technology into their programs.
As these early learners began peppering Spark with questions on topics both relating to the medical interview and far outside the scope of the technology, Spark’s A.I. began using what the platform was taught during the supervised learning to move towards “unsupervised learning.”
The AI would seek out patterns in Spark’s initial training to teach itself to respond to questions the platform had not seen before – the true beginning of Spark’s machine learning capabilities. With new users continuing to come on board, soon after PCS Conversation Engine’s dataset of questions reached the tens of thousands. As Spark is a cloud-based software, these improvements to the conversation engine were something that every user of Spark benefits from.
“That’s really one of the coolest things about Spark, we could have a user on the other side of the world interviewing a Spark patient and phrase a question in a way we’ve not seen before. Our team routinely reviews those logs and maps the new question to the correct topic,” said Nick Stoick, VP of Sales for PCS. “Let’s then say that a user here in the United States phrases a question in a very similar manner, well now the AI has been trained on a similar question and will know how to accurately respond. So literally every day Spark is learning and getting smarter as more people continue to use it.”
This is not just Spark learning new ways to ask questions related to the history of present illness, but also a way to broaden the horizons for learning about other topics as well.
“Last year, we were really fortunate to have worked with the Colleges and Institutes of Canada on the Virtu-WIL project. As part of that project, we developed a number of virtual patients with scenario authors across Canada,” said Stoick. “One of those authors was interested in the topic of diversity and inclusion. We thought that was a great idea, but it was also something that we’ve never trained Spark on before.”
Similar to the landscape when Spark was first launched, he explained that PCS had to start from scratch so-to-speak to begin training around these topics and how the PCS team thought learners might discuss them. When this was internally tested, the results weren’t perfect, but every new interaction PCS had with the new patient acted as a training point for the AI. Now, every Spark user has benefitted from this endeavor as the platform now understands a multitude of topics related to diversity and inclusion, Stoick explained.
More About PCS
For some companies, healthcare simulation is not just a market, but a calling that cannot be silenced. This is true for PCS – a company that started in 2015 and has helped revolutionize patient safety. PCS’s AI-driven patient simulation technology includes virtual patients, wearable simulation, and smart manikins. The company developed the first patient simulator with a fully integrated cloud-based AV recording and debriefing system, enabling clinical simulation centers to digitally debrief sessions with ease.
PCS’s story began back in the 1990s when the founders engaged on a journey around patient healthcare simulation and standardized patient methodologies. The team has since worked with some of the most innovative experts in healthcare simulation as they developed disruptive products.
Overall, the PCS.ai platform is the most advanced cloud-based simulation platform. Users can manage all PCS scenarios and medical simulations through an intuitive, easy-to-use web interface. They can author new patient scenarios, automatically assess learning objectives, and enable learner-driven simulation for their learners as well. The PCS Communication engine leverages the latest advances in machine learning and artificial intelligence to constantly improve. Best-in-class speech recognition and speech synthesis combined with custom natural language processing neural networks.