We are using the term “adaptive training capabilities” too broadly.

Simulation platforms have used adaptive capabilities for decades in different ways based on the actions taken by the participant during the simulation event. Take video based de-escalation and tactics training, for example. If a participant communicates, de-escalates, escalates, or uses force during the scenario, the operator determines what happens next based on the participant’s performance. The scenario changes because of something the learner did. That's adaptive capability.

Many newer virtual simulation systems do the same thing through software. Instead of an operator manually selecting the next branch, the logic is built into the platform. If the learner does this, then the simulation does that. 

What I think is worth questioning is whether some of what is now being described as new adaptive capability is actually older branching logic being repackaged under newer terminology.

I am not arguing that this type of adaptation has no value because it does. If a learner makes a mistake during a medical scenario, the simulated patient should respond to that error. If a participant fails to de-escalate a threat, that threat may escalate. If a learner takes the correct action, the scenario should respond differently than if they took the wrong one. That interaction is part of what makes simulation effective in the first place.

Where I think the conversation needs to move to is how we define adaptive capability beyond the individual scenario. There is a difference between a simulation changing conditions because of what the learner just did and a training system using what that learner has demonstrated over time to influence what they should train next.

This type of capability is where I believe simulation is beginning to evolve. Performance telemetry, observational data, objectives completed, decisions, communication, and other data points can begin contributing to a larger picture of the learner. Performance across multiple events begin building an understanding of demonstrated functional skillsets. The system can then use that information to influence the follow up training experience instead of treating every scenario as an isolated event.

Both approaches are adaptive but the difference is what the adaptation is focused on. One is primarily focused on changing the simulation experience in real time. The other is focused on using performance across multiple training events to influence the learner’s progression over time.

Custom scenario creation tools are already trying to solve part of this problem by giving instructors the ability to build training experiences that better fit their curriculum. The challenge is that most of the work still falls on the instructor. Developing a new 15–20 minute scenario can take roughly 6–8 hours when you account for reviewing the applicable doctrine, aligning TLOs and ELOs, structuring a scenario that fits the curriculum, and THEN going into the simulation platform to actually build and add that scenario to the library. The creation tool helps with the final step, but the instructor is still doing most of the work required to determine what the learner should experience next. This is another area where adaptive capability can evolve. If the system already understands what the learner has demonstrated, the training objectives they are working toward, and where performance is beginning to break down, adaptive generative capabilities should be able to shorten that process by helping structure the next training experience around those needs.

I believe we are at the beginning of dynamic scenario generation capabilities being baked into simulation platforms because future platforms will need to do both. They will continue adapting situations in real time based on learner action or inaction, while also aggregating performance data, measuring that performance against defined training objectives, and using the results to influence what the individual or team experiences next.

The fifteenth scenario should not be another scenario in the library. It should be influenced by what the learner or team demonstrated during the previous fourteen.

As more companies describe their platforms as having adaptive training capabilities, I think it is worth looking past the terminology and asking what the system is actually doing. Is it changing the conditions of the current scenario? Outside of the AAR as a resource for the instructor, is it using what the learner has demonstrated over time to influence what comes next? Are there generative capabilities that can help close the gap and standardize parts of the instructor workflow?