Cognitive Assistant for Training Cost Estimators : Acquisition Innovation Research Center , February 28 , 2025
February 28, 2025
Acquisition Innovation Research Center
Cognitive Assistant for Training Cost Estimators : Acquisition Innovation Research Center , February 28 , 2025
February 28, 2025
Acquisition Innovation Research Center
From the report: "The research team has developed a cognitive assistant to support the training of cost estimators in the Department of Defense (DoD). A Cognitive Assistant (CA) is defined here as an Artificial Intelligence (AI) tool, usually with a natural language interface, that augments human intellect in a specific task by retrieving and processing relevant information from multiple information sources and providing it to the user at the right time. It also has the capability to learn and adapt to the user and the problem at hand.
Cost estimation is a complex iterative process consisting of various steps: gathering the required information, selecting an overall strategy and one or more existing models, developing new models if needed (including calibration and validation), performing the estimate, and conducting sensitivity analyses as appropriate. There are challenges for beginner cost estimators in each of those steps, including dealing with incomplete datasets, appropriately assessing the performance of new models, projecting beyond historical ranges of validity, adequately reporting the level of uncertainty around a point estimate, understanding how to use joint cost-schedule distributions, etc. More experienced cost estimators may also struggle to learn the specifics of a new commodity (e.g., the cost drivers, relative orders of magnitude, etc.).
Currently, the training of new cost estimators is done primarily through traditional instruction in live classrooms, and thus it is a time-consuming process. Traditional instruction typically implies reduced opportunities for hands-on learning opportunities, which areknown to improve learning. This type of instruction is also not tailored to each individual, so the pace can be too fast for some trainees and too slow for others. The use of CAs can allow for more interactive and tailored instruction for eachindividual and area, as demonstrated with intelligent tutoring systems in other areas of education (Corbett et al., 1997).
The idea of using AI tools to enhance the learning of trainees is not new and has been studied for decades (Ong & Ramachandran, 2003). However, in the DoD Acquisition context, we are still in the early stages of incorporating advanced AI tools into workflows and, in particular, CAs have not been adopted yet as training tools. Previous attempts to adopt this technology in the workplace failed because of a combination of insufficient performance of the underlying machine learning (ML) models and lack of familiarity of the users with this mode of interaction. With CAs now being ubiquitous in our daily lives, and the significant recent advances we have seen in machine learning, the time is now ripe for infusion of this technology in the workplace."
Authors - Selva, DanielSubjects
Authors
Publishers
Format
Related Resources
