Organizations across various industries are beginning to realize the power of big data to inform decision making and make predictions. Nowhere is the opportunity more rich than the virtual training space.
Imagine if you could sit in a flight simulator and within just a few minutes, be able to accurately predict your aptitude for becoming a successful pilot. Months, years, and many dollars in training investment could be saved per student — or for that matter, per organization — when critical intelligence is applied at the beginning of the process. But AI-powered models aren’t just useful in identifying learning potential, they can also offer organizations the ability to rapidly modify training in real time to correct undesired behaviour and tailor education to each learner; finally, true individualized learning is actually possible. Unpacking the breadth of this opportunity involves understanding what data should be measured, how it’s collected and processed, and how it can be used to support and enable learning.
We can begin taking models used successfully by global companies such as Boeing to measure and predict the performance of thousands of mechanical parts in real time, then apply similar principles to datasets collected from training simulators to draw intelligent conclusions and assign appropriate reactions.
Step one is to establish what data should be collected. Drawing again from the flight simulator example, we could track body movement through motion sensors on the seat, track which buttons and controls were touched and when, while also scanning the user’s eye movement and direction. All of these data points can be corresponded to the movements of the ‘airplane’ and what impacts those microscopic movements have on the outcome of the mission. This can offer tremendous intelligence when it comes to predicting safety-critical errors (a mistake that would result in a life-threatening event) and mission-critical errors (a mistake that would result in failure to complete the task, such as dropping a package at a specified location).
We would begin by running highly experienced pilots (10k hours or more) through a helicopter simulator, or run the navy’s best operations room teams through a battle simulation, to establish benchmarks for performance and also to collect data on the peripheral behaviours (eye and body movements, reaction time, how buttons and controls are handled and when) that were present. Now we have our high bar of ‘good’ performance and can measure against that.
Where the real intelligence begins to emerge is after hundreds of students have moved through the training and patterns can be identified that are predictive of success or failure. Students who demonstrate undesirable behaviours can be corrected sooner in the process, and students who are lacking aptitude in critical areas that can’t be learned such as spatial reasoning and the ability to turn a 2d display into 3d situational awareness, can be redirected into a discipline where they will be able to succeed and offer more value to the organization.
The collection, synthesis, analysis and application of the data follows a similar flow to what we’ve used successfully in the past for militaries and companies like Boeing. We begin with the sensory trackers which export data to binary or .csv file. From there the data goes to engineers or data scientists for cleaning (removing missing rows, ensuring consistency). After that, data would be stored via SQL/no SQL database for access by systems analysts, data analysts and other stakeholders. From the database, models can be built for the prediction discussed: when X happens, what is the % likelihood of an ultimate Y outcome? And to make this intelligence accessible and actionable, visualizations are generated. Training dashboards can be generated, information systems can be mobilized to put those predictions to use.
Reports that are generated can inform instructors who is performing well, who is struggling and what specifically seems to be causing the struggle (is it a vision problem, reaction time, a physical movement?). Training modules can be added or taken away in real time, such as when a critical error is performed and a student needs to ‘unlearn’ a certain behaviour.
Sometimes though, it’s not the users who are making the mistakes. Data analytics can also reveal flaws in the lesson — if a high number of users are failing at specific instruction, it could indicate an issue with the phrasing of the instruction vs an issue with the user.
Returning to the experienced pilots used for the benchmarking—once a model is properly trained, it may have the ability to make the original 10K + master pilots or the fleet’s best bridge team even better by optimizing everyone’s best practices into an “ideal” practice for them to aspire to.
Aviation and naval training are only a couple of examples of where data analytics can offer an exciting and highly efficient opportunity to enhance training outcomes. Applications in medicine, engineering, sports and skilled trades could save students and institutions volumes of time and expense while also improving outcomes for every learner.
Abdu Radi, PhD
Dr. Radi is an experienced data scientist with a background in physics, chemistry and educational psychology. Most recently he worked for Boeing in Vancouver, where he was responsible for the statistical analysis and modeling work stream for the KC135 and KC10 to determine the working condition of each plane part. As RaceRocks’ Senior Data Scientist, Dr. Radi applies his expertise in machine learning, AI, and data analysis to RaceRocks’ innovative flight simulation and navy training methods.