Tools like ChatGPT, Bing Chat and Jasper Chat have supercharged the effectiveness of scenario-based training programs like never before — not only by dramatically increasing the number of scenarios presented to the learner, but also improving the relevancy and variability in virtual learning environments. In this article, we’ll review several keys to consider when integrating generative artificial intelligence (AI) into scenario-based health care training programs.
Realistic Scenarios in High Frequency and Complexity
The most effective online training programs run through multiple simulated scenarios where learners can practice new skills while receiving feedback on the job. Prior to generative AI tools being widely available, these virtual encounters were programmed using complex decision trees to determine the next question posed to the learner. The amount of effort needed to create and program these scenarios limited how many potential decisions and actions a learner could actually encounter.
These models become even more powerful when trained on specific data sets as opposed to those LLMs that rely solely on the Internet for their source of information. We created one of those very specific data sets to train the LLM powering our training modules. The result is that the AI responses given during each scenario are not only adaptable, reactive and unique, but they’re also based on information and data specific to the setting which can make them much more applicable and relevant.
Make Scenarios Self-Sustaining
With each encounter, the LLM gathers new content which is added to the dataset. Unlike fixed scenario-based training that requires new programming to add new encounters, generative AI “learns” from each encounter and adds that experience to its dataset. It’s best to avoid using humanized terms like “learn” but this is one way to think of the new information added to the LLM. It’s not that the model actually learns something in the human sense, but rather the LLM acquires new information that can be accessed when it generates its next output. Consequently, the conclusions it draws grows increasingly complex with each interaction.
However, it’s critical that to preserve fidelity and sustainability of the module, a team of subject matter and technical experts review interactions and make changes to improve and refine the LLM.
Allow Perpetuity and Specificity
The LLM on which generative AI tools are built on can be licensed for countless of applications from writing social media copy to creating recipes, which can be helpful for most people’s needs but are distinctly non-specific in health care training or other specialized training programs. What generative AI learns from the general use it engages each day is probably not going to add specific knowledge needed for specialized training applications.
To further compound this issue, many training programs rely on the same LLM. While that may be sufficient in these early months, an LLM specific to the training situation that gathers new data from each exercise is going to be better and more relevant to its use case. Just as human variation in real-life scenarios can never be predicted, over time the LLM can gain complexity and depth which results in fresh scenarios being presented every time.
Enable Real-Time Adaptations
From a development standpoint, it’s easier to insert a new situation to the LLM at the data level that can feed directly into the learner’s interface. Imagine how beneficial this capability would’ve been during the COVID-19 pandemic. For example, a mental health awareness training program based on scenarios limited by a decision-tree would’ve required new programming for work protocols. However, with generative AI running the scenarios, those newest processes and situations can be inserted into the LLM in a fraction of the time and with greater fidelity.
Enhance Data Optimization
Generative AI can generate and test a number of scenarios to determine if they’re effectively producing the desired learning outcomes. AI recognizes patterns and can adjust to those patterns faster than humans can. Consequently, new learner scenarios are synthesized, programmed and uploaded faster than ever before the data are optimized.
Data optimization can benefit individual learners, too. AI can personalize scenarios based on individual responses, even recognize weaknesses and present scenarios to retrain those weaknesses. This is the kind of personalized training that would take place in a clinical setting, where a mentor or training supervisor would come to know each trainee and make slight adjustments to the questions or repeat scenarios as needed.
Adding Voice to Complete the Virtual Experience
By now, most people are aware of popular LLMs like ChatGPT and Bard and have given some thought to the possibilities of using generative AI to write articles, answer questions, generate social content or perform any number of helpful tasks. Of far greater value, however, is the ability for LLMs to be trained for specific applications and then voice programmed. This is where we sit with respect to generative AI in training. In health care training programs, the LLM can respond to voice prompts, and this makes the training even more realistic.
Learners can engage in realistic patient interviews simulated to respond like a human. Each response is unique and yet highly realistic to the situation. Imagine being able to ask the virtual patient any question and receive a unique and intelligible response. Furthermore, that interaction is no longer bound by a finite number of interactions. The learner can ask, pivot, clarify and re-ask the same question, probing the virtual patient in ways that feel very realistic. Then, when those responses from the virtual patient are entered into an electronic health record (EHR) it can continue to feed the LLM with new data.
This cycle creates a self-sustaining learning environment unlike anything yet seen before. And, when put into practice, even greater numbers of people can be trained very effectively to fill critically needed positions in the allied health care field.
