Testing is a vital part of education in the modern world. They are crucial for evaluating student learning and achievement, and their uses range from deciding on college admissions to diagnosing knowledge gaps in the classroom. Yet developing good tests is slow and expensive, largely because writing questions that are psychometrically valid (the question actually measures what it should be measuring) is labor-intensive and usually requires domain-experts. My research seeks to understand how we can use to AI to automatically create valid test questions, without needing to rely on humans, to reduce the workload educators face when it comes to assessment.

If you've ever taken a personality, intelligence, or creativity assessment, you've probably engaged with a psychometric test, which is designed to measure something about you that can't be directly observed yet still influences your behavior. Psychometrics is foundational to the psychological and learning sciences, as understanding human behavior requires understanding how properties as intangible as personality or morality (to name a few) influence how people respond to different situations. My interests in this space are divided into two parts. First, I'm interested in using AI to predict human psychometric properties of test questions, so that we can understand how different populations of humans are likely to respond to a test. Second, I'm interested in whether constructs measured in humans also exist in AI, and how to measure them if so. For example, if someone says that ChatGPT has a "pleasant" personality, is this behavior driven by a construct equivalent to human personality that also exists in AI? Or is it driven by something else entirely?

A recent report from the World Economic Forum ranked creativity as being one of the most sought-after skills by employers. Improving creativity in both humans and AI requires that we have psychometrically valid ways of measuring creativity at scale. My work seeks to develop automated creativity scoring systems using AI, such that the responses to creativity tests can be scored with minimal human oversight. I'm also interested in understanding what biases exist in AI scoring systems, both as a means of improving the fairness of their use, and as a way to understand how AI judges the creativity of products on a behavioral or cognitive level.