Artificial Intelligence Quotient (AIQ)

Abstract

We introduce the concept of Artificial Intelligence Quotient (AIQ)—defined as a person’s ability to use AI to perform a wide variety of tasks—and provide evidence for its existence using five studies (archival, lab, and online) across different AIs and samples. Study 1 (an 18-year global dataset of human+AI chess tournaments) and Study 2 (a three-wave longitudinal study of human+AI renju games) show that individuals have stable human+AI performance over time (controlling for human’s own capability and AI’s capability), suggesting the existence of a stable human+AI capability. Study 3 shows that a general AIQ factor can be statistically extracted from individuals’ performance on a variety of tasks completed with ChatGPT, a more general AI tool. Besides replicating Study 3’s findings in larger samples, Study 4 and Study 5 (preregistered) show that the extracted AIQ factor has both concurrent validity and prospective validity. Regarding concurrent validity, the extracted AIQ factor can predict human+AI performance on a new task using the same AI (ChatGPT) on the same day. Regarding prospective validity, the extracted AIQ factor can predict human+AI performance on other new tasks using different AIs (renju AI or Gemini) in the future. Across studies, we ascertain the unique explanatory power of AIQ by controlling for individual’s IQ, social intelligence (SQ), AI literacy (knowledge about AI), and/or computer literacy. We also explored potential correlates of AIQ (e.g., personality traits, previous AI use, and demographics). Together, our findings suggest that AIQ exists and is measurable. By establishing this new type of intelligence (AIQ), we shed light on individual differences in the ability to use AI, which is increasingly important for individuals, organizations, and society.

Ähnliche Arbeiten