Embeddable measurement models for cognitive psychometrics
| Agency | National Science Foundation |
| Panel | Methods, Measurement, and Statistics |
| Location | University of California, Irvine |
| Start date | September 2026 |
| End date | August 2029 |
| Budget | $ 365,562.00 |
| Agency code | 2549818 |
Abstract
In this project, the research team makes use of recent progress in cognitive science. They use mathematical representations of how humans normally make quick decisions in order to measure the mental sharpness and decision preferences of many people at once. Usually, testing a person's focus or decision-making takes a lot of time in a laboratory, and often requires a powerful computer to analyze the data. Supported by top-of-the-line AI agents, the team is creating fast, simple versions of these tests that can be run on normal smartphones and tablets. The project includes multiple sub-projects to apply these faster methods to bring large benefits to society. For example, the researchers use them with phone apps made to detect early signs of memory loss or dementia, which may help to treat Alzheimer's Disease early in some people. There is also a sub-project that helps scientists more easily keep track of the mental health of large numbers of children over time. Additionally, a collaboration with the Air Force uses the quick focus tests to make sure pilots are not too tired to fly safely.
The main problem to be studied is that computer models used to understand human choices are often too slow for real-world applications. Advanced models of decision-making are great at explaining how people think because they can separate mental speed from other factors that determine how fast someone acts (for example, how cautious they are). However, these models require so much computer power that they cannot be used on large data sets, and they can never be used in time-sensitive situations. The goal of this research is to create highly efficient "proxy" models. These proxy models capture the most important parts of the original cognitive models but take much less computer power to run. To do this, the team uses a mathematical technique that lets them focus on high-level patterns in the data rather than a highly detail-oriented technique that was used before. To understand these high-level patterns, the researchers will train artificial neural networks to learn and copy the steps of the original models. Once built, these fast proxy models can replace the detailed models, leading to faster measurement and measurement at large scale.