This research strikes me as hot garbage :/ In particular there are dozens of studies like this, which compare to AI with "humans" without adequately addressing sample bias:
Human participants (N = 151) were recruited via Prolific online data collection platform in exchange for monetary compensation of $8.00. Participants were limited to having a reported approval rating above 97%, were proficient English speakers, and were born/resided in the USA. Average total response time for completing the survey was 34.66 minutes.
Asking a bunch of random people to fill out tedious "creative" writing prompts for $15/hr is not a good way to measure humans. Baffling that the discussion didn't even mention this. But you see this all the time in AI research, and I am sure this headline will be repeated uncritically by many people.