Excited that our study on individual differences in human discernment of generative-AI texts (masquerading as social media news posts) is out in #ScientificReports. Main findings and some examples you can test yourself on in this thread ... www.nature.com/articles/s41... - ThreadSky

mindimager.bsky.social • 64 days ago

Excited that our study on individual differences in human discernment of generative-AI texts (masquerading as social media news posts) is out in #ScientificReports. Main findings and some examples you can test yourself on in this thread ... https://www.nature.com/articles/s41598-024-76218-y

Comments

amiyake.bsky.social•62 days ago

Very interesting work. Any follow-up plan, esp. to better understand the nature of individual differences? I was surprised that verbal ability of some sort was not assessed and wonder to what extent the fluid intelligence result can be explained by verbal ability (or other correlates of FI).

mindimager.bsky.social•59 days ago

Great question. We’re following up by testing how detection might change with development. We do have some verbal processing measures that we can use a covariates of interest. We’re also trying our hand at a novel intervention to improve detection accuracy, though I’m skeptical it will work.

mindimager.bsky.social•64 days ago

Finding 1: Human discernment of text-based generative AI (is it human or AI?) is just slightly above chance guessing, but some people are strikingly good at making the determination (> 80%).

mindimager.bsky.social•64 days ago

Finding 2: What makes the especially good human "detectors" so good, in a nutshell, is their high fluid intelligence. Greater tendency to empathize, on the other hand, doesn't matter. Human intelligence is, at least currently, a defense against being fooled by artificial intelligence.

mindimager.bsky.social•64 days ago

Finding 3: More time on social media and using smartphones increases exposure to AI, which makes us more likely to misattribute human origin to AI material. So, people who use digital media more excessively also have a harder time telling the difference between what's human and what's AI.

mindimager.bsky.social•64 days ago

Finding 4: People who can't discern human from AI-generated materials are, unfortunately, also more likely to spread information online by sharing, which means that the people who just can't tell the difference are also the one's most likely to be propagating online information. Damn!

mindimager.bsky.social•64 days ago

Here are some examples to try (reply if you think its "human" or "AI", answers at bottom of thread):

mindimager.bsky.social•64 days ago

E1. "Researchers unveil groundbreaking findings that shed light on previously unknown factors affecting storm intensity and trajectory. This discovery will greatly enhance forecasting accuracy, helping protect communities from the devastating impacts of these natural phenomena."

Comments

Posting Rules

Reply