Skimming the paper, they're not arguing that the computer can detect if someone is trustworthy, they're arguing they can detect is something is *perceived* as trustworthy.

If so, why couldn’t the author be bothered to mention that in what feels like a hundred tweets?

Grateful it's not technical people publishing ethical tone-deaf papers with ML this time Less disturbing to see it from subjective humanities people like psychologists rather than real scientists

It seems I was wrong and the paper actually says the opposite of what everyone, including me, assumed