Temporary Permanence

Temporary Permanence

Lukas Vacula's personal website and blog.

06 Jan 24

Commenting on a Paper from First Monday about Generative AI

One of the 400 RSS feeds in my FreshRSS instance is an open-access journal that describes itself as "solely devoted to the internet": First Monday. Reviewing this month's articles had one that was particularly interesting for me: Why do people use ChatGPT? Exploring user motivations for generative conversational AI". I encourage you to read it.

This post will be some comments on the paper (mostly just the points I found interesting).

  • There were 197 participants in the study. Okay, this one isn't interesting but it will help contextualize the numbers below.
  • 58% of participants reported using ChatGPT weekly or more, but only 4% used it daily. I initially thought the difference would be because they're using it for work (and thus not on the weekends), but using it for personal use is more popular (82% for personal use versus 38% for work).
  • The most popular theme for ChatGPT use was productivity at 55%. That tracks with my priors. Interest in the technology was second at 51%, then a big drop down to only 20% for fun and amusement.
  • Some people used it to come up with dinner suggestions (no complaints here!). Others used it to set up diet plans (scary!) or "a place to address mental health issues" (also scary!).
  • 5 of the participants reported "high reliability or trustworthiness of the information it offered". I'm reminded of a question someone asked of where the idea that generative AI was reliable for information came from. Why do people assume it is good at being correct and not confidently wrong? That was one of the first things that came to mind when I first heard of it.
  • More people (24) reported using it as a support tool for writing, such as to re-word their own sentences, than as a purely generative tool (10). This feels slightly better to me than if pure generation was more popular.
    • This difference was narrower in software developement. 17 used it for generation and 14 for debugging/problem-solving. I think I feel less good about this.
  • Two of the summarized uses that the authors list in the discussion - alleviating the burden of decision making and dealing with information overload - feel like they're connected. What is the burden of decision making but an information overload related to a particular choice?