bioRxivJClub
Individual differences in artificial neural networks capture individual differences in human behavior
bioRxiv · · Preprint · Open access
Fung, Murty + 1 more
Abstract ↗AI summary
The abstract is read at the publisher; the summary is JClub's.
Sixty instances of the same ANN architectures on digit recognition show substantial individual differences in accuracy, confidence, and response time, mapping to 60 humans with mapping strength often near the human-to-human benchmark.
- Why it matters: The gap is whether ANNs can capture individual variability seen in human behavior, which is needed to probe mechanisms behind why people differ.
- What they did: The study trained and tested 60 ANN instances from three architectures on digit recognition, then compared each ANN’s behavioral metrics to 60 humans; it also tested a 10-choice blurry object recognition task.
- The result: ANN–human mapping generalized across metrics (accuracy alignment predicted confidence and RT alignment for the same individual), and extended to the 10-choice task, though mapping was generally less robust than human–human.
The findingWhy it mattersWhat they didThe result
- Open access