Human-like individual differences emerge from random weight initializations in neural networks

Herrick Fung, Apurva Ratan Murty, Dobromir Rahnev

Abstract: Much of AI research targets the behavior of an *average* human, a focus that traces to Turing's imitation game. Yet, no two human individuals behave exactly alike. In this study, we show that artificial neural networks (ANNs) trained with different random initializations exhibit substantial individual differences that resemble those in humans. Using a large dataset (N = 60) of human responses (accuracy, confidence, & response time) in a digit recognition task, we trained multiple instances of three ANN architectures on the same task, creating as many ANN instances as human subjects. We found that these ANN instances vary significantly from one another. Critically, ANN instances showed consistent variation in their alignment with specific human subjects. This consistency in alignment between ANN instances and humans extended across behavioral metrics, indicating that an ANN instance mimicking an individual on one metric also does so on others. Finally, we showed that leveraging these alignments improves predictions of individual human responses. Our findings highlight the potential of ANNs to capture human variability, opening new directions to develop models that go beyond aligning the *average* human and instead aligning the idiosyncratic behavior of specific *individuals*.

Length: short paper (up to 4 pages)

Domain: methods

Author List Check: The author list is correctly ordered and I understand that additions and removals will not be allowed after the abstract submission deadline.

Anonymization Check: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and URLs that point to identifying information.

Submission Number: 8