Sebastian Karcher
I've mentioned this paper simulating survey responses using GPT-3 to several people #IASSIST2023 . It isn't uncontroversial, but important to be aware of:
https://doi.org/10.1017/pan.2023.2
ungated: https://arxiv.org/pdf/2209.06899.pdf
Out of One, Many: Using Language Models to Simulate Human Samples | Political Analysis | Cambridge Core

Out of One, Many: Using Language Models to Simulate Human Samples - Volume 31 Issue 3

Cambridge Core
Jun 2, 2023 at 2:14pmWeb
Show threadDatendealerin 🎃☕Jun 2, 2023
@adam42smith 😳 And here we go...
Show threadSebastian KarcherJun 2, 2023
@Datendealerin yeah. It'll be virtually undetectable. I'm quite curious if GPT 4 is good enough to successfully mimic semi-structured interviews. Data provenance will take on a whole new dimension
Show threadGuillaume PasquierJun 2, 2023
@adam42smith @Datendealerin
Show threadDatendealerin 🎃☕Jun 2, 2023
@gap @adam42smith Time for a career change 🙈
Show threadDatendealerin 🎃☕Jun 2, 2023

@adam42smith GESIS had a project a while ago on how to detect falsified interviews. From what I remember, they couldn't make the interviewers fake the interviews in an experimental session 😉

This calls for a whole new research topic.